Codicology is the term for the study of old manuscripts (= MSS) and the use of them to produce authoritative texts. The term is derived from the Latin 'codex,' which is the word for manuscripts bound roughly in the modern sense. In producing an edition of a text, other sources of information, including papyri and inscriptions, may be used; in practice, however, most ancient texts are preserved in handwritten codices (the plural of codex) produced from about a hundred years before the fall of the western Roman empire until about the dawn of printing. What follows is a highly streamlined explanation of the process of drawing up a critical edition, derived from my experience with classical texts.

A classical text is likely to exist in multiple copies spread over Europe and beyond. The MSS of most classical texts are known and have been catalogued. The first thing to do, therefore, is find out where MSS of your text live by consulting the standard catalogues, and to order microfilms of them. In the rare (but scarcely unheard-of) event that your text exists in a hitherto unknown or uncatalogued copy, you will need to study the physical object on which it is written, as well as the handwriting. (It will eventually be necessary to check the physical MSS corresponding to all of the microfilms you used, too: the photographs, in black and white, invariably have problems with focus, centering, etc., and you'll have a short-list of folios to check in person.)

The material (vellum vs. paper) can give an indication of age (paper was introduced later than vellum). Paper may exhibit watermarks applied by the manufacturer; these have been studied pretty thoroughly, and serve as a good terminus post quem for the production of the codex. Ink compositions changed over time, and the color of the ink may thus help date the MS. In addition, paleography (i.e., handwriting styles and modes of abbreviation) changed over time, and is also a good indicator of the age of the copy. Naturally, the copyist or an early owner of the MS may have written something which will give a date and other information--in the best case, you will find the copyist's name, and be able to date him and find out where he worked.

Codices have a habit of being frankenstein monsters, cobbled together out of the parts of several copying sprees or from the viscera of multiple earlier codices. A medieval collector might have fancied a certain era of ancient history, or astronomy, and caused a florilegium of several works to be created by detaching interesting parts of several codices and binding them within new covers. Attentive study of the writing material (size, hand, lines per page, ink colors, portions of the relevant works preserved, etc.) may just pinpoint an earlier codex from which your piece of text was chosen. See an excellent example of a physical and paleographical description of a codex here.

Once you have dates (and other relevant information) for all of your MSS, it is time to begin looking at the text. At this point your labor switches from codicology proper to textual criticism, that is, the critical evaluation of the surviving copies of a text. The best idea is to choose the earliest, most complete copy of your work and create a transcription in modern orthography. Your copy may or may not be the best surviving, but that is not important yet. You are assembling a skeleton against which you are going to collate (= systematically compare) the readings of your other MSS. If your preliminary inspection suggests that one MS offers a more reliable text than others you look over, then use it (see the indented paragraph below).

If you know a lot about your MSS, you may in fact know which ones are descendents of others. In your first round, you can set aside descendents of extant MSS because they are not independent witnesses--in theory (if not always in practice), they cannot be better than the MS from which they were copied.

A text surviving in several (or even many) MSS will quickly exhibit some distinctive patterns. At points, some MSS will agree with your first MS, while others will disagree. Over time, you will find that the "variant readings" cluster regularly. MSS A, B, and C may regularly agree against MSS D and E. This indicates (if all are independent witnesses) that they are descended from two earlier (but presumably now lost) MSS; we might call the ancestor of ABC "a", and that of DE "b". In this way, you break down the MSS into families. Often, families differ strongly in quality: "a" might have been a very early copy of the archetype (= the "mitochondrial mother" of all the extant MSS, if you take my meaning), perhaps in the 5th century; "b" might have been copied only a few years before its first extant descendent (say, in the 13th century). "a" (= ABC) will usually give better readings than "b" (= DE).

Once you have your families (and you may well find more than the two families I posit here), and you are in a position to decide which family has the best pedigree, you will probably want to adopt the readings of that best family in preference to the others.

All copyists, being frail human beings, introduced errors into their copies if the latter were of nontrivial length. Entropy is the name of the game here, and the more stages of copying a text has been through to reach a certain exemplar, the more cumulative errors will exist. In our hypothetical example from above, certainly A, B, and C will differ among themselves--just not as regularly as they do as a group from D and E. It should be possible to isolate the best MS of the best family to use as a base text.

(Aside: I cheated above by avoiding a step too complex to explain there. The way to choose which MS to use as your base text is to do a little survey--collating maybe 10,000 words from all of the independent witnesses (in sections where the text survives intact, of course). The family relationships will sort themselves out pretty quickly, and you will probably be able to discover the oldest MS in the best family.)

Collation is time-consuming, eye-destroying work, and many editors of a century or more ago settled for a collation of the independent witnesses to a text. (In fact, they were often prominent scholars who used indentured labor--er, graduate students--to travel and perform the collations. For understandable reasons, many of these collations are unreliable, both because of graduate students' inexperience and perhaps their resentment at or disinterest in another man's project into which they had been dragooned.) In the best of all possible worlds, however, the editor will have a look at everything, including, at a later stage, the witnesses dependent upon the independent ones already collated.

At this stage (or even earlier), interesting similarities may be found between a MS of one family and one in another. G, a direct copy of E (and hence, in our family "b"), may show some strong similarities with "a", and in particular with some interesting readings unique to A. From your codicological study you may know that G was copied from E on Crete, for example, and that A was in another monastery on the island. In this case, you could be pretty sure that the copyist of G almost certainly had not only E on hand, but had also borrowed A, occasionally taking from the latter readings he thought better. This is called contamination, and it can complicate things a lot. Occasionally, a late MS may exhibit extremely good readings not seen elsewhere, even though it is from the weaker family. Chances are, it was copied in the presence of a now-lost MS from a family better than any that still exists. That's the way of it.

How do you choose the "better reading"? The time-honored principle is lectio difficilior preferenda est: the harder reading is to be preferred. (See that writeup for details.)

In the end, you will have a text which, at any given point, has one or more variant readings. You want to print the best text possible, which means you want to follow the best MS of the best family except when another family's superior reading forces you to bring the latter into the text; put all other variants at the bottom of the page in a separate section keyed to textual line numbers. This section is called the apparatus criticus, and besides offering "all the data" in the name of honesty and completeness, it is a tacit admission that you (being a frail human being yourself) may not always have identified the best reading, and you offer your readers the opportunity to think about it for themselves. It will also be necessary to introduce modern punctuation and possibly capitalization conventions. Greek texts will need a frustrating check of all the accents and breathings.

An edition to which this sort of critical thinking has been applied, and which is fortified with critical scholarly apparatus, is called a "critical edition." In ancient studies, the best available are generally the Oxford Classical Texts series, and the slightly superior Bibliotheca Teubneriana. In New Testament studies, the critical edition to get is the Novum Testamentum Graece edited by Nestlé and Aland (now in a 27th edition), published by the American Bible Society.

Maas, P. 1958. Textual Criticism.
West, M.L. 1973. Textual Criticism and Editorial Technique Applicable to Greek and Latin Texts.