It has been recognised that the only rational procedure is to begin by
determining in what relation the copies stand to each other. For this
purpose we adopt as our starting-point the incontrovertible axiom that
all the copies which contain the same mistakes in the same passages must
have been either copied from each other or all derived from a copy
containing those mistakes. It is inconceivable that several copyists,
independently reproducing an original free from errors, should all
introduce exactly the same errors; identity of errors attests community
of origin. We shall cast aside without scruple all the copies derived
from a single manuscript which has been preserved. Evidently they can
have no value beyond what is possessed by their common source; if they
differ from it, it can only be in virtue of new errors; it would be
waste of time to study their variations. Having eliminated these, we
have before us none but independent copies, which have been made
directly from the archetype, or secondary copies whose source (a copy
taken directly from the archetype) has been lost. In order to group the
secondary copies into _families_, each of which shall represent what is
substantially the same tradition, we again have recourse to the
comparison of errors. By this method we can generally draw up without
too much trouble a complete genealogical table (_stemma codicum_) of the
preserved copies, which will bring out very clearly their relative
importance. This is not the place to discuss the difficult cases where,
in consequence of too great a number of intermediaries having been lost,
or from ancient copyists having arbitrarily blended the texts of
different traditions, the operation becomes extremely laborious or
impracticable. Besides, in these extreme cases there is no new method
involved: the comparison of corresponding passages is a powerful
instrument, but it is the only one which criticism has at its disposal
for this task.

When the genealogical tree of the manuscripts has been drawn up, we
endeavour to restore the text of the archetype by comparing the
different traditions. If these agree and give a satisfactory text, there
is no difficulty. If they differ, we decide between them. If they
accidentally agree in giving a defective text, we have recourse to
conjectural emendation, as if there were only one copy.

It is, theoretically, much more advantageous to have several independent
copies of a lost original than to