Abstract : This thesis explores the problems raised by the aggregation of entities into a global, collective level, an old problem encountered in many fields of science. We work on three projects related to the aggregation problem in social systems, using tools derived from statistical physics, and more generally quantitative tools. The first project focus on a paradigmatic model of the emergence of puzzling macroscopic behavior from simple individual rules, Schelling's segregation model. We hence propose an analytical resolution of this model and we studied analytically and via simulations the effect of several forms of cooperation between individual agents on the collective behavior. These questions are tackled in a mutually beneficial way for both economics and physics. The second project is based on the exploration of huge databases on scientific literature. We hence produce several 'science maps' representing the fields of complex systems (its internal structure and coherence being analysed through the references used by ~140000 relevant articles) and the research carried out in a scientific institution such as the ENS de Lyon. Finally, the third project deals with the elaboration of models of social phenomena based on natural sciences tools but sociologically grounded. We hence present the elaboration process of a model built with a team of sociologists. We then propose an opinion model specifically designed to explore a single question: the existence of lasting structure from non lasting entities.