Some dimeric proteins first fold and then dimerize (three-state dimers) while others first dimerize and then fold (two-state dimers). Within the framework of a minimal lattice model, we can distinguish between sequences following one or the other mechanism on the basis of the distribution of the ground state energy between bulk and interface contacts. The topology of contacts is very different for the bulk than for the interface: while the bulk displays a rich network of interactions, the dimer interface is built up of a set of essentially independent contacts. Consequently, the two sets of interactions play very different roles both, in the folding and in the evolutionary history of the protein. Three-state dimers, where a large fraction of energy is concentrated in few contacts buried in the bulk, and where the relative contact energy of interface contacts is considerably smaller than that associated with bulk contacts, fold according to a hierarchical pathway controlled by local elementary structures, as also happens in the folding of single-domain monomeric proteins. On the other hand, two-state dimers display a relative contact energy of interface contacts, which is larger than the corresponding quantity associated with the bulk. In this case, the assembly of the interface stabilizes the system and leads the two chains to fold. The specific properties of three-state dimers acquired through evolution are expected to be more robust than those of two-state dimers; a fact that has consequences on proteins connected with viral diseases.