Historically they arrived first, ranging from simple majority fraction to information entropy and entropy related methods, to full-blown statistical estimation of the mutability of residues leading to the observed set of sequences. Such methods work well in detecting the folding core of a protein, the catalytic site of an enzyme, and somewhat less reliably, the protein-protein interfaces shared by all homologues. Their performance is affected more strongly by the preprocessing stage, then by the choice of method itself. The specialization of duplicated genes is the necessary condition for their parallel existence, and the methods to detect it on the protein level followed shortly. Several major ways of treating this problem have been put forth, differing mainly in the way they handle. The first issue has been dealt with by taking the classification as an input, by using the similarity tree as the classification generator, or by adopting a midway solution in which the tree is provided by the application, but the relevant division into subtrees is decided on by the user. In this work, we would like to put some emphasis on the way an evolutionary model is built into a specificity scoring function. As an example, a popularly quoted evolutionary trace method, ET, in its original formulation assumes that a functionally important position will be completely conserved in each of the compared groups of sequences, albeit as a different amino acid type. If the groups in question are paralogous, this becomes a very strict model of evolution, in which even after the duplication and specialization event, each gene maintains the same degree of evolutionary pressure at each site. This model appears in the literature in several forms. Conversely, mutual information requires that each group of orthologues adopts a set of evolutionary constraints that are systematically different from those of all other groups, irrespective of the degree of conservation within each group. However, it mirrors “conservatism-of-conservatism” in conditioning the expected behavior in one group, on the behavior in another. Recently, ever more voices appear in the literature, pointing out that the evolutionary behavior in paralogous groups may be completely unrelated. Variously termed “type I functional divergence” or “SB431542 301836-41-9 heterotachy”, this type of behavior has been discussed in genetics literature for at least a decade, and used increasingly in detection of family specific positions on a nucleotide or peptide sequence. Finding the “type I – type II” terminology somewhat lacking in descriptive power, we use the term “determinants” for the positions that are conserved in one group, but evolve at various rates across paralogues, and “discriminants” for the positions that vary at comparable low rates across all groups. A determinant position, then, is a property of a single group, while a discriminant is a property of the family as a whole.