Abstract In a similar way in which the folding of single‐domain proteins provides an important test in the study of self‐organization, the folding of homodimers constitutes a basic challenge in the quest for the mechanisms that are the basis of biological recognition. Dimerization is studied by following the evolution of two identical 20‐letter amino acid chains within the framework of a lattice model and using Monte Carlo simulations. It is found that when design (evolution pressure) selects few, strongly interacting (conserved) amino acids to control the process, a three‐state folding scenario follows, where the monomers first fold forming the halves of the eventual dimeric interface independently of each other, and then dimerize (“lock and key” kind of association). On the other hand, if design distributes the control of the folding process on a large number of (conserved) amino acids, a two‐state folding scenario ensues, where dimerization takes place at the beginning of the process, resulting in an “induced type” of association. Making use of conservation patterns of families of analogous dimers, it is possible to compare the model predictions with the behavior of real proteins. It is found that theory provides an overall account of the experimental findings. Proteins 2002;49:82–94. © 2002 Wiley‐Liss, Inc.