Canons of Criticism

Contents: Introduction *Outline of the Canons *External Critical Rules *Internal Critical Rules *How to Use the Canons of Criticism *Footnotes


Although detailed methods vary, there are really only two ways to edita Bible text. One is to print a text based on some sort of external control(the Textus Receptus, the text found in the majority of manuscripts, thetext found in B/03). This may be useful, and may fit the publisher's assumptions,but it hardly constitutes editing. Its's more an exercise in reading anillegible hand.

The only other way is some form of eclecticism-- picking and choosing between readings.And, unless one is content to print a chaotic text, choosing between readingsrequires some sort of guidelines. These guidelines are the "canonsof criticism."

Outline of the Canons

Different editors have listed different rules, and applied them in differentways. Some have listed dozens of criteria,[*1]others only a handful. No matter how many rules they list, all fall intoone of two categories: Internal criteria (pertaining to the logicof readings) and External criteria (pertaining to the manuscriptscontaining the readings). Thus there are only two fundamental canons:



All other canons -- no matter how numerous or how detailed -- are simplycorollaries or specific examples ofthese two rules. (The only so-called "critical method" which does notoperate on this basis exception is the Byzantine Priority technique which simplycounts noses. As no editor has ever published an edition based solely on thiscriterion, we can ignore it.)

Still, as any mathematician will tell you, the general rule may be pretty,but it's usually much easier to apply specific formulae.[*2]The sections which follow describe some of the better-known rules for criticismthat various scholars have used. Note that, since each is a specific caseof a general rule, they should only be applied in the appropriate situation.The discussion tries to describe the situations in which which each ruleapplies. I've also tried to list who first proposed the rule, or who popularizedit.[*3]

External Critical Rules (pertaining to manuscripts)

That reading is best which is supported by the best manuscripts.This was the fundamental tenet of Hort, and has been followed by many others-- including even Lagrange and Weiss, who in theory explicitly rejected it. Thisis a good rule if all the best manuscripts support a single reading (i.e.if all the leading manuscripts of all the early text-types agree), butshould not be applied by itself if there is disagreement among the text-types.Still, this rule may be the final arbiter if all other criteria fail. Also,to apply this rule, one must have a precise definition of the "best"manuscripts. Unless one is Hort, and prepared to follow B/03 blindly, thisrule can be hard to apply.

The geographically superior reading is best. I deliberately state thiscriterion vaguely, because geography has been used in various ways by variouscritics. The usual sense used in New Testament criticism is Streeter's, who argued that thereading supported by the most diverse sets of "local texts" is best. I.e.his criterion is That reading is best which is supported by the most geographicallydiverse manuscripts. That is, if reading X is supported by manuscriptsfrom Rome, Carthage, and Alexandria, while reading Y is supported onlyby witnesses from Byzantium, reading X is to be preferred. The rule goes back toBengel, but was strongly reinforced and restated by Streeter. All thingsbeing equal, this is a good rule, but there are two limitations. First,good readings may be preserved in almost any text (e.g. there aremany instances where scholars read the text of B/03, perhaps supportedby a papyrus or two, against all comers). Second, this rule can only beapplied if one truly knows the provenance of manuscripts. (For additionaldetail, see the entry on Local Texts.)
There is, however, another rule based on geography, more commonly encounteredin classical criticism but with some application to New Testament criticism,especially in studies of text-types and smaller textual groupings: Themore remote reading is best. That is, isolated sites are more likely topreserve good readings, because manuscripts preserved there are more likelyto be free from generations of errors and editorial work. This criterion, ofcourse, cuts two ways: While a remote site will not develop the errors of thetexts of the major centres, it is more likely to preserve any peculiar errorsof its own. Remote texts may well be older (that is, preserve the readings ofan older archetype); they are not automatically more accurate.

That reading is best which is supported by the earliest manuscripts.This was the basis of Lachman's text; he used only the earliest manuscripts.Today, it finds support from Aland (who has referred to the papyri as "theoriginal [text]") and also Philip Wesley Comfort, who has the tendencyto treat all papyrus-supported readings as accurate. It is, of course,true that the papyri are valuable witnesses, and that the support of earlymanuscripts increases the likelihood that a reading is original. But othercriteria must take precedence. This is a rule of last resort, not a ruleof first resort.

That reading is best which is supported by the most manuscripts.This is, of course, the negation of the theory of Hort, whose primary purposewas to dethrone the Textus Receptus. Although this rule has some modernsupporters (e.g. Hodges, Robinson), it is generally rejected. Certainlythose with scientific training will not be impressed with "MajorityRule." Modern eclectics of all types generally feel that, at best,this rule should be avoided until all other means of decision have failed.(Note: This is not saying that the reading of the Byzantine text is wrong.It's just that it's only one text type; adding more and more witnessesto the type does not change that fact.)

That reading is best which goes against the habitual practice ofparticular manuscripts. So, for instance, P75 and B have been accusedof having exceptionally short texts -- of omitting (by design or chance)many pronouns and other "unnecessary" words. So where P75 and Bhave a long reading, their testimony bears particular weight. By contrast,D is considered to include many interpolations and additions. Where,therefore, it has a short reading, the short reading is consideredespecially probable. (This is the theory, e.g., behind the so-called"Western Non-Interpolations.")Note that this rule can onlybe applied if the habits of a particular manuscript are truly known.

The reading of the first hand is best. This requires a good dealof caution in application, but it is Miroslav Marcovich's first rule for editingpatristic texts: "Wherever possible, establish and follow the first handof a manuscript." This is presumably because the first hand is older andis less likely to be using corrupt sources. But we know that there areexceptions, the most obvious being 424, where the corrected text is much moreimportant than the base text.

That reading is best which endured longest in the tradition.That is, a reading which is found in manuscripts from (say) the ninth tofifteenth centuries is superior to one found only in the fourth and fifthcenturies. This criterion, offered by Burgon, has recently been re-statedby Pickering.[*4]Moderns apparently apply this rule in some cases (e.g. Eph. 1:1, wheremost scholars include the words "In Ephesus," even though themanuscript evidence against them -- P46ℵB 6 424** 1739 -- is verystrong). I know of no eclectic scholar who states the rule, though, andmost of the time they actively reject its dictates; see, for example,2 Cor. 12:9, where ℵ**A** D** K L 0243 33 330 1739 Byz (sixththrough sixteenth centuries) read "my power is perfected in weakness,"while P46-vidℵ*A*vid B D* F G latt sa (thirdthrough ninth centuries) omit "my." The fact that every trulyearly witness omits "my," and that these witnesses come fromthree different text-types, counts for nothing when using this criterion.Therefore scholars reject the rule; all editions since Tischendorf (saveHodges & Farstad and Pierpont & Robinson) have omitted "my."

Great diversity of readings often indicates early corruption andperhaps editorial work. This principle, in use since the last century,has recently been firmly restated by Kurt and Barbara Aland. The difficulty,of course, lies in figuring out which reading is original when confrontedby a wide variety. It should be noted, however, that in the case of such corruption,the original may be found in manuscripts which otherwise would not be foundreliable. A good example is 1 Thes. 3:2, where the best-attested readingwould appear to beδιακονοντουθεου(ℵ A P 424** bo arm). Of the half-dozendifferent readings here, however, the best appears to beσυνεργοντουθεου,supported only by D* 33 d Ambrosiaster.

The continuous reading is best. Maurice Robinson, who strongly supportsthis rule, states it in full as follows: "In any extended passage wheremultiple sequential significant variant units occur, those MSS which offerstrong support in less problematic variant units are more likely to be correctin the more problematic units if such MSS retain their group supportwithout serious fragmentation of or deviation from such group." This ruleonly applies in groups of three or more points of variation. Let us considerthe simplest example, of three sets of variants (call them A, B, and C).Suppose you can clearly decide thecorrect reading in A and C, but are not certain about the reading in B. In thatcase, the manuscripts which are correct in A and C are likely to be correct inB as well. The logic is that scribes are basically careful. They transcribe accuratelyif they can, but one or another condition may cause them to slip. If a scribeis transcribing most variants in a passage accurately, chances are that heor she will have done equally well in variants where we cannot assure his orher accuracy.
This rule is difficult to demonstrate in practice, because of the greatdiversity of methods of criticism. A reading which one critic considersuncertain may seem quite assured to another critic. And critics do notagree on textual groupings, either. It may not be possible to offer anexample of this rule which would be accepted by all critics. Certainly Iknow of none. So I will offer a hypothetical example, not because I likeusing artificial examples but because I'd rather have something that clearlydemonstrates the point.
Consider the following passage, based loosely on John 11:25. The variantsare enclosed in curly brackets. We will assume that each reading is supportedby a certain collection of text-types: A=Alexandrian, B=Byzantine,C=Caesarean, W=Western. (Note that one need not accept the existence of anyof these types; any set of groupings would be equally meaningful here):
απεκριθη{Ιησους AB |κυριος Ιησους CW}{και ειπεν BC | omit AW},{εγω BW | omit AC}ειμι η αναστσιςκαι η ζωη
Most critics would agree, based on either internal or externalevidence, that the short readingΙησουςis correct in the first variant. And stylistic considerations dictatethat the third variant should read ειμι,not εγω ειμι. But whatabout the inclusion/omission of και ειπεν?One reading is shorter and more direct, the other more typicalof Johannine usage. So internal evidence fails us, at least at a casual glance. In such a case,we turn to the criterion of the continuous reading. In this case, theAlexandrian text is clearly correct in the first and third readings. Chancesare, then, that it is correct in the second reading also; we should omitκαι ειπεν.
The danger with this criterion lies in over-applying it. This is notthe same as the rule that the best manuscript/text-type is best. (ThoughMaurice Robinson believes that this lesser rule generalizes to that greaterprinciple.) This is a local principle, applying to relatively short passages.Moreover, it is a secondary rule, applying only to uncertainvariants in the context of variants which are secure.

That reading found in the majority of early text-types is best.OK, a personal opinion here: This is the rule. The whole story.If you have three early text-types (call them "Ptolemaic,""Romanesque," and "Cilician,") and two of themattest to a particular reading, doesn't it stand to reason that themajority of the text-types -- all of which go back to the original --is more likely to be right unless there is some other explanation forhow they came to be corrupted? Curiously, no one seems to have appliedthis rule on a consistent basis. The problem, of course, lies indetermining what is a text-typeand which of them are early. This is an area that doesn't get nearlyenough attention -- which in turn means that this most basic andobvious and objective of rules is not stated, and rarely applied;no one is willing to do the work to apply it!

Internal Critical Rules (pertaining to the natureof variants)

The shorter reading is best (Lectio brevior praeferenda).This rule is found in most manuals, beginning with Griesbach, and certainlyhas its place. There were scribes who liked glosses, and there were scribeswho would always prefer the longer reading (on the principle that it wasbetter to have an extraneous word in scripture than to risk leaving somethingout). However, this rule must be applied with extreme caution (as Griesbachhimself noted, adding exceptions for scribal errors and for minor omissionsthat do not affect the sense). The most common sorts of scribal errors(haplography) result in a shortening of the text. Also, there is a strongtendency among copyists to omit short words. (These first two errors areboth characteristic of ℵ,for example.) In addition, there were scribes(the scribe of P45 is perhaps the most extreme) who freely shortened thetext. Finally, despite Boismard, the short reading should not be adoptedbased only on arguments from silence (Boismard adopts a number of shortreadings in John on the grounds that patristic sources omit the words.This is not good evidence; the phrases in question may simply not havebeen relevant to the commentator's argument). Therefore the rule of the"shortest reading" should be applied only if the manuscriptswith the short reading are reliable and if there is no evident reason whyscribes might have deliberately or accidentally shortened the text. Asa general rule, if a scribe makes a deliberate change, it will usuallyresult in a longer text; if a scribe makes an error, it will more oftenresult in a shorter text.
At this point it might be worthwhile to quote G. D. Kilpatrick: "Thereare passages where reasons can be found for preferring the longer text andthere are others where we can find reasons for preferring the shorter. There isa third category where there does not seem to be any reason for deciding oneway or the other. How do we decide between longer and shorter readings in thisthird category? On reflection we do not seem able to find any good reason forthinking that the maxim lectio brevior potior really holds good."("The Greek New Testament of Today and the Textus Receptus,"in Anderson & Barclay, The NewTestament in Historical and Contemporary Perspective," 1965, p. 196.)
Still, there are cases where this rule is accurate, though usually for other reasons thansimple brevity. An obvious example of the use of this rule is the several additions of"fasting" with "prayer," e.g. in 1 Cor. 7:5 (Mark9:29 is also an example of this type, although it is perhaps a questionableinstance since the external support for "and fasting" is very strong,and the words are found in all manuscripts which insert the sentence in Matthew. Thisimplies that those who added the words to Matthew must have known them inMark).

The hardest reading is best (Difficilior lectio potioror Proclivi scriptioni praestat ardua). First offered by Bengel(for whom it was the basic rule), this is a good criterion; scribeswere generally more likely to make texts simpler rather than harder. Butsome caution must be applied; scribes were capable of making errors thatled to prodigiously difficult readings. (A good example of this is thepeculiar readings that litter P66.) One should prefer theharder reading only when it is adequately attested and does not appearto be the result of error. Or, perhaps, the rule should be rephrased:Among readings which are possible, the hardest reading is tobe preferred.

The reading most in accord with the author's style is best. Thisis a two-edged sword, since copyists were perfectly capable of conforminga peculiar passage to an author's style. Take the Gospel of John. Thereare dozens of instances of the phrase "Amen, amen, I say to you."Suppose the author had, in one instance, left out an "Amen"?Would this reading have survived in the tradition? Perhaps not. And ifit had survived in one part of the tradition, might not an editor beinclined to reject it? If applied with caution, however, this rule can bevery useful; it often allows us, e.g., to choose between verb forms (sincemost authors have a peculiar pattern of verb usage.) Of course, the usageof the author must be known very well.
This is sometimes stated in Greek, Πλάτωνα ἐκ Πλάτωνος.

The middle reading is best. This rule is rarely found in thetextbooks, even though Griesbach had a form of it -- though biologicalstemmatists have found it very important and useful. It obviously only appliesin cases where there are three (or more) readings. If there are three readings,X, Y, and Z, and a simple change will convert X to Y, and Y to Z, but nosimple change will convert X to Z or vice versa, then Y is the middle reading(the one that could have given rise directly to the others), and is tobe preferred. Of course, this only applies where X, Y, and Z all have earlyattestation. If one of the readings is late, then it could be a tertiarycorruption.
An example of the use of this rule occurs in 2 Pet. 2:13. Here P72ℵA* C? 33 81 436 614 630 1505 2344 Byz readαπαταις,A** B Ψ 623 1243 1611 vg readαγαπαις,and 322 323 945 (1241) 1739 1881 readαγνοιαις.Most editors explain away αγαπαιςas an assimilation to Jude 12. However, there are good arguments for its originality.In addition, it is the middle reading; bothαπαταις andαγνοιαις could have arisen directly fromαγαπαις but could not have arisenfrom each other. Since all three readings are early, andαγαπαις is the middle reading,it is to be preferred.

The reading which could most easily have given rise to the otherreadings is best. This approximates Tischendorf's formulation of thegeneral rule "That reading is best which best explains the others."It is a direct corollary of the basicrule, and has much the same force as the preceding rule and applies in all thesame cases and more.

The reading which could not have arisen from lectionary use is best.Many continuous-text manuscripts were marked for lectionary use. Oftenthis meant adding lectionary introductions, and often these introductionscrept into the text (the praxapostolos 1799, for instance, is litteredwith lectionary incipits). If a reading might have arisen as the result of thiserror, it is probably to be avoided. Compare the following rule:

The reading which is counter to ecclesiastical usage is best.Offered by Eberhard Nestle, this applies mostly to passages found in thelectionary. It also argues against readings such as "Amen" at the end ofepistles: With the exception of James (where "Amen" is found in 614 14481505 1852 (2492) 2495 t hark pc),at least one uncial witness attests to "Amen" at the endof every New Testament epistle. However, the editors of UBS/GNTaccept the word only at the end of Galatians, Jude, and -- in brackets-- 2 Peter.

The disharmonious reading is best. This rule is usually appliedin the gospels, where assimilation of parallels is common. If one readingmatches the text of another gospel, and the other reading does not, thenthe assumption is that the unique reading is best. (Von Soden noted a specialinstance of this: All things being equal, scribes tended to assimilateto Matthew as the "strongest" of the gospels. If no other ruleresolves a variant involving parallels, The reading which does not matchMatthew is best.) This is a good rule, but must be applied with caution.As Colwell has shown, the most common sort of assimilation is assimilationto the immediate context. Also, scribes would sometimes assimilate to other,unrelated sources (e.g. hymns or other writings that sounded similar tothe scripture being copied). So this rule should really be altered to read...
The less familiar reading is best. That is, if one reading is whatyou would expect a scribe to write, and the other is unusual or surprising,the latter is probably the correct reading. This is what Hort called "TranscriptionalProbability." The only problem is guessing what was going on in thescribe's head as he wrote....
We can illustrate this with an example from the LXX. Consider Ezek.38:13. The Hebrew text refers to "Tarshish." The translators of LXX glossedthis to the more familiar "Carchedon" (Carthage). But the scribe of A wasconfused even by that, and converted it to "Chalcedon." We see thisidentical error in some classical texts, from the period when every Byzantinescribe knew the Council of Chalcedon but when Carthage was a forgotten cityin the west: In Aristophanes, Knights 1303, manuscripts R VΨ refer toCarchedonians/Carthaginians, butΓ2and some scholia mention Chalcedonians.

The reading which best fits the context or the author's theologyis best. If we were absolutely sure of how the author thought, thiswould be a good rule. As it is, it is awfully subjective....

The reading which has the truest sense is best. Hort said thatthe best readings are those which, on the surface, don't make sense, butwhich, on reflection, show themselves more reasonable. Hence this criterion.Perhaps the best example of its application is the reading of UBS/GNTin 2 Cor. 5:3, where (following D* (F G) a d f** g) that text reads "ifindeed, when we take it off, we will not be found naked." Allother witnesses, starting with P46, read "...when we put it on,we will not be found naked." The UBS editors accept the reading"take it off" on the grounds that the other reading simply doesn't make sense.

The reading which avoids Atticism is best. With the Attic Greekrevival of the early Christian centuries, Attic forms began to be usedafter some centuries of disuetitude. Kilpatrick, in particular, calledattention to Atticising tendencies. The caution with this rule is to determinewhat is a truly Attic reading and what is legitimate koine.Parallel to this rule are the three which follow:

The reading which is characteristic of Hellenistic usage is best.Since the koine used a number of unclassical and uncouth forms,later scribes with more classical education might be tempted to correctsuch "barbarisms." This is another of the stylistic criteriaof Kilpatrick and Elliot. Fee, on the other hand, denies it; scribes seemoften to have conformed readings to the koine and Septuagint idiom.

The reading which resembles Semitic usage is best. Since mostof the New Testament authors were native speakers of Aramaic, they wouldtend to use Semitic idiom in violation of Greek usage. Copyists, as nativeGreeks, might be expected to correct such readings. This is again the argumentof the thoroughgoing eclectic school (compare the preceding rules), andagain there are those who argue that scribes would be more likely to preferSeptuagintal usage.

Parallel to the two preceding is The reading which is less like theSeptuagint is best. This is another of those tricky rules, though. It'scertainly true that some scribes would tend to conform to the Septuagint. Butthis has even more than the usual complications. It must be remembered, forinstance, that most copies of LXX were made by Christians, and they might oftenconform LXX to the New Testament usage more familiar to them -- meaning thatthe harmonization, rather than being in the NT, is in LXX! And then, too,NT authors often deliberately used LXX language which scribes might mis-copy.

That reading which seems to preserve an ungrammatical form is best.A trivial example is Mark 6:29 (ηλθαν/ηλθον), where firstand second aorist stems are interchanged. Most applications of this rule are toequally trivial matters -- although sometimes they may reveal something aboutthe scribe who produced the manuscripts.

The metrically superior reading is best.Much of the Bible is poetical, and even much that is not is still metrical -- justas much of Shakespeare is in blank verse, because blank verse sounds likejust plain old spoken English, but is smoother. In non-poetic passages, think ofit as an "accounting" meter, to fit the rhythms of the language or hold tosome particular pattern of content. If a metrical pattern is detected,the variant that conforms to this pattern is more likely to be original. Somestudents have seen very regular patterns in the text which can even be used asa sort of guideline for the use of this canon.

If one reading appears to be an intentional correction, the readingwhich invited such a correction is best. Alternately, That readingwhich is most likely to have suffered change by copyists is best. Proposedby Tischendorf. This is fundamentally the same as preferring the harderreading. If a reading calls out for correction, of course some scribeswill correct it. They are hardly likely to deliberately create a reading which requiressuch correction. An obvious example is Mark 1:2. HereℵB (D) L Δ(Θ)(f1) 33 565 (700) 892 1241 it arm geo read "Asis written in Isaiah the Prophet," while A W f13 579Byz read "As is written in the prophets." The citationwhich follows is, of course, from several sources, only one of which isIsaiah. While it is possible that scribes corrected "in the prophets"to "in Isaiah the Prophet" based on parallels (since so manyNT citations are from Isaiah), it is much more likely that scribes corrected"in Isaiah the prophet" to "in the prophets" to eliminatethe errant reference.

The reading which could have given rise to the others accidentallyis best. Or, as P. Kyle McCarter puts it, Look first for theunconscious error. This is a very important rule in Old Testamentcriticism, where independent witnesses are few. It is less applicablein the New Testament, where witnesses are frequent and where errorsof spelling or dittography are less likely to give rise to a meaningfulvariant. However, if one reading could have given rise to another byan accidental error (e.g. by omitting a doubled letter or a shortword or syllable), that reading is clearly to be preferred.

The reading which is susceptible to a heterodox interpretation isbest. This rule does not often apply, but when it does, it is important.A reading which lessens the dignity of Christ, for instance, is usuallypreferable (unless it is supported only by highly questionable sources).Examples of readings where this criterion applies include:

If there were any doubt about the operation of this rule (and thereshouldn't be, because we see Origen casting out the "Jesus Barabbas"reading because he didn't like its implications), we can see its operationin action in classical texts. In Odyssey XIII.158, the manuscriptsread μεγα δε, whichcauses Zeus to say to Poseidon, in effect, "Go ahead! Flatten thosePhaeacians for being kind and hospitable to visitors." This was sotroubling that Aristophanes of Byzantium claimed the proper reading musthave been μηδε, whichmakes Zeus reluctantly allow a limited punishment rather than addingrefinements to Poseidon's capricious cruelty. This sort of theologicaltampering continues today; the Richard Lattimore translation of theOdyssey accepts this reading!

The reading which contains unfamiliar words is best. Offeredby Metzger (following Griesbach) in conjunction with some other observationsabout scribes. The change from the unfamiliar to the familiarcan happen (it happens very frequently in oral tradition),but is not as likely as it sounds. (Consider the wordεπιουσιον in theLord's Prayer. No one to this day knows what it means with certainty --but scribes never tried to change it!) If a scribe knows a word, he willnot object to copying it. If the word is unfamiliar, how is the scribeto know what word to replace it with? In applying this criterion, it isbest to know the peculiar habits of a particular manuscript.

If, in a variant reading, one reading is subject to different meaningsdepending on word division, that reading is best. I don't rememberwhere I came across this, and I can't cite an example by chapter and verse; itcertainly doesn't come up often. (Souter gives two examples,1 Tim. 3:16,ομολογουμενωςor ομολογουμεν ωςand 2 Tim. 2:17,γαγγραινα orγαγγρα ινα.But neither of these involve variants in the actual text.) But I recalla variant something like this. Suppose some manuscripts readΟΙΔΑΜΕΝand othersΚΑΙΟΙΔΑ.Since the former could be read as eitherοιδαμεν (one word) orοιδα μεν (two words),and so is ambiguous, it is preferable.

If a reading is a conflation of two shorter readings, the shorterreadings are best (though the correct reading must be decided on othergrounds). This rule, used by Hort to demolish the Textus Receptus,is good as far as it goes, but conflate readings are actually very rare.The best-known example is probably Luke 24:53. Here P75ℵ B C*L sin cop geo read "blessing God," D a b e ff2 read "praisingGod," and the remaining witnesses (including A C** W Θ f1f13 33 892 1241 Byz) read "praising and blessing God."Since the reading "praising and blessing God" is a conflationof the Alexandrian reading "blessing" and the "Western"reading "praising;" it is to be rejected. As between "blessing"and "praising," the decision must be made on other grounds. (Mostscholars would prefer "blessing," both because it is the Alexandrianreading and because it is more presumptuous -- how dare people "bless"God? But that decision is made based on other rules. The rule against conflatereadings only allows us to eliminate the conflate reading.)
Another good example is Matthew 10:3,where the readings "Lebbaeus called Thaddeus" and "Thaddeuscalled Lebbaeus" are obviously attempts to combine the Alexandrianreading Thaddeus and the "Western" reading Lebbaeus.
In using this rule, one must also be careful to try to reconstruct howthe conflation came about. For example, in Mark 15:39there is a possible conflation, since the various readings areεξεπνευσεν,ουτωςεξεπνευσεν,κραξαςεξεπνευσεν,andουτωςκραξαςεξεπνευσεν.I have argued elsewhere that the manuscript evidence here indicates that the "conflate"readingουτωςκραξαςεξεπνευσενis most likely original.

The true reading is best. This is offered by Wordsworth and White,who stated it as, "The true reading wins out in the end." Althoughthis might be interpreted as an argument for the majority text, or thelate medieval text, that is not how Wordsworth & White used it. Howthis rule is to be applied must therefore be left as an exercise for thereader. Bentley's version was simply to apply "common sense."Apparently early textual critics had never heard American politicians talkabout what is sensible....

The reading which is contrary to the habits of the scribe is best.This can be applied to individual manuscripts, in which case it is hardlya canon of criticism, but is very useful in assessing the habits of a particularscribe. For example, D/05 has been accused of being anti-Jewish and anti-Feminine.If, therefore, it has a reading that is pro-Jewish or pro-Feminine, thatreading is likely to predate the prejudiced handling of D (compare theexamples in the next item). Similarly, if P75 is given (as many believeit is) to omitting pronouns, and somewhere it has a pronoun not found inother Alexandrian witnesses, the evidence for the longer reading is strengthenedbecause P75 went against its habit, implying that the reading comes fromits exemplar. This criterion, although appealed to by eclectics of allsorts, is apparently particularly dear to Elliot and the thoroughgoingeclecticists. If applied at a level above that of individual manuscripts,though, it says little more than "study what Hort called 'transcriptionalprobability.'"

That reading which violates the prejudice of scribes is best.This may sound like the previous rule rehashed. It isn't, exactly, althoughit also applies first and foremost to individual manuscripts. This hasbeen pointed up by Ehrman and others in connection with the Christian prejudiceagainst Jews. So, for example, if one reading is anti-Jewish and the otheris neutral, the neutral reading is to be preferred. (Ehrman offers John4:22 as an example, where some versional witnesses read "salvationis from Judea" rather than "...from theJews.")[5]Also falling in this category is the treatment of Prisca the wife of Aquila.Her name occurs six times in conjucnction with his.In four of these instances (Acts 18:18, 26,Rom. 16:3, 2 Tim. 4:19), her name appears first in the best witnesses (sheis listed second in Acts 18:2, 1 Cor. 16:19). But in Acts 18:26 (D 11751739 Byz), some manuscripts demote her to the position after Aquila. Inaddition, in Rom. 16:3 (81 223 365 630 876 1505 1881** ful* pm),1 Cor. 16:19 (C D F G 81 Byz a d ful tol), 2 Tim. 4:19 (206 223 323 429436 876 2412 a ful al) the manuscripts listed demote her name from"Prisca" to the diminutive "Priscilla." This could justbe assimilation to the more familiar usage -- but it could be prejudice,too.

Where the same variant occurs in parallel passages, each variantis original somewhere. I have not seen this canon formally stated(and so provided my own statement), but it is used in a number of places(e.g. by the editors of the New English Bible). Three examples may bestexplain the situation:

  1. Matt. 8:28=Mark 5:1=Luke 8:26, Gerasenes/Gadarenes/Gergesenes
  2. Matt. 10:3=Mark3:18 Lebbaeus/Thaddaeus
  3. 2 Pet. 2:13=Jude 12 ΑΓΑΠΑΙΣ / ΑΠΑΤΑΙΣ

In the first instance, the NEB reads Gadarenes in Matthew, Gerasenesin Mark, and Gergesenes in Luke. In the second, it has Lebbaeus inMatthew and Thaddaeus in Mark.
One must take great care in applying this criterion, however. The NEB approachis probably wrong, at least in the case of the Lebbaeus/Thaddaeus variants.The key observation has to do with text-types. In both Matthew and Mark,the Alexandrian text reads Thaddaeus, while the "Western"text reads Lebbaeus. (The Byzantine text conflates in Matthew.) Inother words, this is not a case where the two gospels had differentreadings but where two different traditions had different names forthis apostle. We are not trying to decide which name to use in whichbook; rather, we must decide between the two names overall. Whichevername is original in one book is original in the other.
This is not to say that this criterion is without value. One mustsimply be very careful not to use it where it is not relevant.

If similar variants occur in several places, the reading morestrongly attested in the later points of variation is best. Or,as Maurice Robinson phrases it, "If a particular type of phrasingrecurs several times within a book, but in a form rarer than that normallyused by the writer, scribes would be tempted to correct such a readingto standard form at its earlier occurrences, but not in its later occurrences."This rule apparently goes back to Wordsworth and White.

As for what it means, it means that if a scribe is confrontedwith a particular reading -- especially one which seems infelicitous oratypical of the author -- he is likely to correct it the first few timeshe sees it. After seeing it a few times, he is likely to give in --either due to fatigue or as a result of saying something like, "Well,he's said it that way three times now; I guess he meant it."

We in fact see some instances of this in Jerome's work, though in histranslation activity rather than in his copying; early in theVulgate gospels, he was much more painstaking in conforming the Old Latinto the Greek; later on, if the Old Latin adequately translated the Greek,he didn't worry as much about making sure parallel Greek structurestranslated into parallel Latin structures. This seems to be a goodrule, in principle. In practice, I can't cite a place where it wouldbe used.

The Parsimonious Explanation is Best. As far as I know, this rule hasnever been used in a manual of textual criticism -- but it is absolutely vital inscience. "Parsimony" is sort of a technical term for Occam's Razor:The simplest explanation is best. To put it another way, The ExplanationRequiring the Fewest Assumptions Is Best. Unneccessary assumptions are theroot of all evil -- at least when seeking knowledge.

It seems to me that this is best applied when dealing with Kurt Aland's"Local Genealogical Method." The idea is to produce the simplest localgenealogy.

I thought of this while having a row with one James Snapp over the endingof Mark. I am sure I will not present this in a way that is fair to him, but it makesa good demonstration of the number of different assumptions one might make toexplain a variant. What follows is only my take on his presentation of the evidence.Note that what we really discussed was a matter of canonicity more than textual originality,but it can be handled as a textual issue.

Snapp's opinion is that "Mark 16:9-20" is probably not from the same authoras the rest of the gospel, but that it was added to the Gospel before it began to circulate.To put it another way, it is not original to the author, but it is an original part of thepublished text.

He accepts that the gospel originally ended at 16:8 (either due to loss of text orbecause the author never managed to finish the gospel). But this was remedied at anearly stage by the addition of the Longer Ending.

Snapp's suggestion is that some later authority disliked the Longer Ending (arguingthat this authority intended to use the ending of John as a better ending for Mark, offeringthe Gospel of Peter as a parallel). So this person excised the Longer Ending and did -- something.This, seemingly, was detected and the interpolation excised. Hence the version of Mark whichends at 16:8. This circulated widely enough that someone felt the need to add an ending. Hencethe creation of the Shorter ending. From this situation -- versions with no ending, with theLonger Ending, with the Shorter Ending, and with the Johannine Ending -- the current mix ofmanuscripts evolved.

So Snapp's assumptions are:
1. Mark came to end, either deliberately or accidentally, at 16:8
2. At a period before the book was widely circulated, 16:9-20 was added. This makesit an original and canonical part of the gospel as published.
3. 16:9-20 was excised so that a different ending (from John?) could be added. (Interestingthat this later editor could slice off the ending at the precise point where the style andcontent seems to change dramatically. But ignore that.) Thisleft two versions in circulation: That with 16:9-20 and that with the Other Ending
4. The forgery of the Other Ending was detected and excised, leaving three endings incirculation: 16:9-20, Other Ending, no ending
5. The Other Ending was lost, leaving in circulation 16:9-20 and the version with no ending
6. In a region where 16:9-20 was unknown, the lack of an ending was felt as a defect andthe shorter ending was added.
7. The surviving endings (longer, shorter, no ending) combined to produce the currentmix of manuscripts.

By comparison, here are the assumptions underlying the UBS assumption that 16:9-20 aresimply an addition:
1. Mark came to end, either deliberately or accidentally, at 16:8
2. The lack of an ending was felt in at least two distinct places, resulting in theindependent creation of two different endings, 16:9-20 and the Shorter Ending
3. The surviving endings (longer, shorter, no ending) combined to produce the currentmix of manuscripts.

Snapp's reconstruction requires seven steps. The UBS version requires three (arguablyfour, if you count the Longer and Shorter Endings as separate creations). Snapp's reconstructionalso involves a version which has been completely lost (except for debatable parallels suchas the Gospel of Peter and Tatian's Diatessaron). Snapp disagrees vigorously with he considersthe list of assumptions, but this is how I slice it; parsimony argues very strongly againstthe assumption of canonicity for 16:9-20.

Like most canons of criticism, the most parsimonious explanation is not guaranteed tobe correct. And it must be examined in light of the textual evidence. But the mere act oftrying to identify one's assumptions, and seeing if any can be dispensed with, has shown itselfto be an immensely powerful tool in the sciences; I suspect it will in textual criticism also.

Finally, never forget Murphy's Law of Textual Criticism: If you canimagine an error, a scribe has probably made it. (For that matter,scribes have made a lot of errors you can't imagine.) To put itanother way: Never underestimate the sleepiness of scribes. Scribeswho work long hours inevitably get tired, and as they reach the close ofthe day their vigilance will wane. (Zuntz thought he observed thisin P46 in Hebrews, and I see signs of it in C3throughout the New Testament. Students of Egyptian texts often found that thescrolls they were reading were much less accurate as they approached theend.) The result can be hilarious errors.Perhaps the most famous is found Luke's genealogy of Jesus (Luke 3:23-38).In codex 109, the genealogy was copied from an exemplar where the genealogywas written in two columns. The scribe of 109 converted this into one --without observing the gap between the columns! As a result, instead ofGod standing at the head of the list, the ancestor of all is Phares andGod is the son of Aram. It is possible that the strange version of theParable of the Two Sons (Matt. 21:28-31) found in D lat is also the resultof such a stupid error. Confronted with two versions of the story (onein which the first son went and the other in which the second did so),a very early "Western" copyist corrected one form part way towardthe other -- and wound up with the absurd conclusion that the son who refusedto work was the one who did his father's bidding! This rule needs alwaysto be kept in mind in assessing criteria such as "the harder reading."

We find another curious example from an Anglo-Norman manuscript of sermonsby Robert de Greatham. Charlton Laird (The Miracle of Language, pp. 185-186)tells this story: "The scribe who copied the manuscript finished a line whichended in a form of peché (sin). Whether or not this particular scribehad some Freudian interest in sin, when he flicked his eyes back to the manuscripthe was copying from he hit upon another peché which was the last wordin the seventh line previous. Accordingly, he copied the same seven lines twice....No two of these lines agree. Here was the same scribe, with the same [original],who copied the same passage twice within a quarter hour, and he does not produceone single line which is identical in both copies. Nor is he consistent in hisown spelling of common words."

Always look to see what errors a scribe could have made!

How to Use the Canons of Criticism

Different scholars apply the canons very differently. Some place mostof the weight on external criteria; others on internal. Some analyse readingsstarting withinternal criteria, others with external. In other words, people have differentrules for using the rules! [*6]

An article such as this cannot, or at least should not, tell you whatto do. But it might be appropriate to describe how some editors approachthe problem.

As the least of all textual critics, I will start with me. I begin bylooking at text-types. If all early text-types (of which there may be asmany as four or five) agree, then I am done. If, however, the early text-typesdisagree, then I shift to examining the variant. If there aremultiple readings, I attempts to construct a local stemma. (In doing so, we shouldnote, the evidence of the number of types is very important. If one type hasa certain reading, and all the others have a different reading, the more commonreading is much more probable.) If a stemma can be constructed successfully,this resolves the variant. If no certain stemma can be constructed, I adoptthe variant supported by the most text-types; if the types are evenly split,and only then, do I turn to the earliest/best type.

Hort's method (as reflected in the edition ofWestcott & Hort)was basically similar, except that he had only two earlytext-types, and one of them (the "Western") was very bad. SoHort frequently was constructing stemma within the Alexandrian text, orsimply setting aside the "Western" reading and adopting the textof B. Hort did not list canons of criticism, although he stressed the roleof "intrinsic probability" (what the authors had written) and"transcriptional probability" (what scribes did with it). Hissummary of the causes and nature of errors is still relevant today.

The Alands stress the importance of "local genealogy" (thestemma of the various texts in a variant).[7]It is interesting to note, however, that their text very much resemblesHort's. In effect, they were bound by manuscripts as much as he was (notehow many of their "Twelve Basic Rules for Textual Criticism,"rather than being true canons of criticism, simply stress the importanceof manuscripts, or are truisms -- e.g. "only one reading canbe original").

Von Soden's approach was genealogical in another sense. He tended towork based on the majority-of-text-types, after making allowances for corruptions(e.g. from Tatian and Marcion) and for harmonizations. His method, whateverits theoretical merits, was badly flawed by his imperfect text-types andhis inadequate knowledge of the sources he blamed for corruptions.

Harry Sturz's proposed approach (which did not result in a completetext) is to print the reading found in the majority of text-types (Alexandrian,Byzantine, "Western"), with little or no attention to internalcriteria. Since the Byzantine text, in the gospels, agrees with the othertwo more often than they agree with each other, his gospel text appearsto be strongly Byzantine.

Also Byzantine are the texts of Hodges &Farstad and Pierpont &Robinson, both of which accept the Byzantine Majority text as originaland apply various criteria to restore that text.

The "rigorous eclectic" school of Kilpatrick and Elliot givesalmost all its attention to internal criteria. Although it is not entirelytrue, as some have charged, that they only use manuscripts as sources ofvariant readings, it is certainly true that they resolve most variantsbased entirely on internal criteria, and will accept readings with minimalmanuscript attestation.

B. Weiss theoretically used techniques similar to those of the "rigorouseclectics," based primarily on internal criteria and with especialfocus on suitable readings and those appropriate to the author's style.In practice, however, he came to rely rather heavily on B as the best manuscript(and so produced a text with significant similarities to Westcott and Hort).

Tischendorf's approach was in some ways similar; most of his criteriawere based on internal evidence (though he stressed that readings neededto be found in old manuscripts). It is not too surprising that the text ofhis eighth edition (his ultimate work)heavily favored his personal discovery,ℵ.

The method used in the first twenty-five editions of theNestle-Alandtext need hardly be discussed here, since it was based exclusively on earlierpublished texts. It was consensus text of Westcott & Hort, Tischendorf,and Weiss (after the third edition).

Lachmann printed the text found in the majority of the early manuscripts.His text therefore fluctuated badly depending on which manuscripts survivedfor a given passage.

So how does one decide what method to use, and which canons to emphasize?Despite the words of Michael Holmes,[*8]that still remains very much up to the reader. Perhaps this piece willgive you a slightly fuller menu to choose from.


1. Von Mästricht's 1711 edition -- arguablythe first to include rules for criticism -- listed forty-three canons!Most of these are not what we would today call "criteria"; theyare observations about (often attacks on) scribes, or methods for decidingwhat is or is not a variant. But they are historically important, sinceboth Wettstein and Bengel were influenced by them.
It should be noted, however, that the first real study of textualcriticism from the modern standpoint is that of Wilhelm Canterin 1566. Syntagma de ratione emendandi scriptores Graecosoutlined many classes of errors, and probably influenced Bengelat least.
The best summary of the history of criteria is probably Eldon J. Epp, "TheEclectic Method in New Testament Textual Criticism: Solution or Symptom,"printed in Eldon J. Epp and Gordon D. Fee, Studies in the Theory andMethod of New Testament Textual Criticism (Studies and Documents 45,Eerdmans, 1993). The extensive section on canons of criticism begins onpage 144. The history shows clearly how much of the theory of criticismgoes back to Bengel; see especially the summary on page 148. [back]

2. If you want an example, consider this: Ilearned to add starting in first grade. Thus I was doing arithmetic, followinga specific rule, when I was six years old. It was not until I was a juniorin college that I was first exposed to what mathematicians call "TheFundamental Theorem of Arithmetic" (that each number has a uniqueprime factorization). Thus I learned the specific rules a decade and ahalf before I learned the general rule. And, to this date, I have neverused the Fundamental Theorem of arithmetic. [back]

3. The list given here is compiled from a varietyof modern manuals, most of which list only the critical canons accepted and used bythat particular author -- if they list canons at all. This list attempts to show all thecanons the various authors use, whether I approve of them or not. The listof works consulted includes Hammond, Metzger (both the Introductionand the Textual Commentary), Vaganay/Amphoux, Kenyon, Aland& Aland, Black, Lake, and Greenlee, as well as a variety of specialstudies, most particularly by Epp and Colwell. I also looked at severalOld Testament commentaries, and of course the book by Pickering cited below.Not all of these books list canons of criticism (indeed, some such as Lakehardly even mention the use of internal criteria); in these cases I havetried to reconstruct from the examples or from miscellaneous comments.It will be noted that some of these rules are closely associated withclassical textual criticism, but thatothers are unique or nearly unique. For example, New Testament criticismdoes not rely upon manuscript stemma to the extent that classical studiesdo. This is largely due to the massive numbers of Biblical manuscripts (amongClassical sources, only Homer is within an order of magnitude of thenumber of NT sources), which make true genealogical studies very difficult.[back]

4. Wilbur N. Pickering, The Identity ofthe New Testament Text (Nelson, 1977), p. 134. On pages 129-138, Pickeringoffers the first modern support for Burgon's seven "Notes of truth"-- criteria by which a reading is determined to be original. These are:

  1. Antiquity, or primitiveness -- which to Pickering means that an originalreading must be found before the Middle Ages (!).
  2. Consent of witnesses, or number ("a reading attested by only afew witnesses is unlikely to be genuine").
  3. Variety of evidence, or Catholicity (witnesses from many differentareas).
  4. Continuity, or Unbroken Tradition ("A reading, to be a seriouscandidate for the original, should be attested throughout the ages of transmission,from beginning to end.... If a reading died out in the fourth or fifthcentury we have the verdict of history against it. If a reading has noattestation before the twelfth century, it is certainly a late invention.")
  5. Respectability of witnesses, or weight. (Note that Pickering, in offeringthis criterion, adds "The oldest manuscripts can be objectively, statisticallyshown to be habitual liars, witnesses of very low character...." Since Pickeringcan be demonstrated to have about as much understanding of statistics as theaverage lungfish, one must wonder how seriously to take his comments here.)
  6. Evidence of the Entire Passage, or Context (referring not to internalevidence but to how reliable a particular manuscript is in a particularsection of the text).
  7. Internal considerations, or reasonableness (Pickering applies thisonly to readings which are "grammatically, logically, geographically,or scientifically impossible," and gives as an example Luke 19:45,where he apparently prefers "The sun was darkened" to "thesun was eclipsed"; Pickering cites four other examples, but in noneof them was I able to determine which reading he preferred and why.)

It will be noted that all of Burgon's "Notes" except #4 (thecanon to which this note refers) are accepted by other textual critics-- but generally applied in very different ways! If Pickering's versionof Burgon's criteria were applied consistently, then the search for "theoriginal text" would be nothing more than an examination of the Kxrecension. Kx is, by Pickering's standard, old (the earliestmanuscript, E/07, dates from the eighth century); it is always the majorityreading (according to Frederik Wisse, The Profile Method for Classifyingand Evaluating Manuscript Evidence, Studies & Documents 44, Eerdmans,1982, 53% of the manuscripts of Luke are Kx at least in part);its sheer bulk ensures its "catholicity," "continuity,"and "weight," and -- by virtue of being Byzantine, and thereforerelatively easy -- its readings are "reasonable." [back]

5. Bart D. Ehrman, "The Text as Window:New Testament Manuscripts and the Social History of Early Christianity,"printed in Bart D. Ehrman and Michael W. Holmes, eds, The Text of theNew Testament in Contemporary Research (Studies and Documents 46, Eerdmans,1995), p. 366. [back]

6. Eldon J. Epp (in "Decision Points inNew Testament Textual Criticism," printed in Epp and Gordon D. Fee,Studies in the Theory and Practice of New Testament Textual Criticism(Studies and Documents 45, Eerdmans, 1993)), pp. 39-42, speaks of "TheCrisis of Criteria," and even goes so far as to describe the presentuse of "reasoned eclecticism" as a "cease-fire" betweenthe proponents of internal and external criteria (p. 40). This obviouslyimplies an earlier state that was nearly a shooting war.... [back]

7. Kurt Aland & Barbara Aland, The Textof the New Testament (translated by Erroll F. Rhodes, 2nd Edition,Eerdmans, 1989), p. 281, item 8 -- and elsewhere. [back]

8. "In short, reasoned eclecticism isnot a passing interim method; it is the only way forward. As long as oursubject matter is, to paraphrase Housman, the human mind and its disobedientservants, the fingers, hopes for a more objective method will remain animpossible dream." Michael W. Holmes, "Reasoned Eclecticism inNew Testament Textual Criticism," printed in Ehrman and Holmes, p.349. [back]