Abstract : We have developed an algorithm to predict residues implied in the folding nucleus with two coupled approaches. Most Interacting Residues (MIR) prediction is coupled with structural analysis methods by fragments as the Tightened End Fragments (TEF) technique and multiple structural alignment. This algorithm have been developed on a proteins benchmark with low sequence identity belonging to immunoglobulin family which is well known. The results are compared to those produced by similar techniques and by experimental approaches. As a result, there is a good agreement between predicted residues and experimental data. For a second part, we have made a bank of biological multimers. The purpose of this bank is to develop a prediction tool of quaternary structure. The bank of multimers is considered valid by the DiMoVo software using Voronoi tessellation to represent proteins. To analyze types of interacting domains, several complexes have been indented in fragments by the TEF method. We have developed an algorithm to encapsulate the fragments into cylinders to characterize the fragments by their radius and length. Then these fragments have been classified on structural criteria and the interactions into the protein-protein interfaces of the classes have been counted. As a result, we observe that the classes are not used in a homogeneous way in the proteins-proteins interactions.