As it is well known, classification problems in Pattern Recognition obey to two different kinds of decision logic: unsupervised and supervised.
In the problems of unsupervised classification one deals with tentative classifications (usually the classes are expressed according tree structures). The consistency of these "a priori" classifications should then be controlled by means of standard univariate and multivariate statistical tests.
The unsupervised decision logic is the type of logic that is commonly used in archaeological problems, where one seeks tentatively patterns and structures in a data set.
We now turn to the supervised decision logic. In contrast with the unsupervised logic, we have here an "a posteriori" decision pattern: we start with a "learning set" (in most archaeological cases it is the result of a "tentative" unsupervised classification) and then we try to classify a new "unknown" set, according standard statistical decision techniques, such as, for instance, linear maximum likelihood, Fisher linear discriminant analysis, Bayes decision logic, etc.