Abstract : This thesis deals with the use of evolutionary algorithms (EA) to solve constraint satisfaction problems (CSP) in finite domains, without any particular specialization nor hybridization. Having presented the CSP and general methods used to solve them (chapters 1 and 2), we present the evolutionary paradigm and its applications (chapters 3 and 4). Then, we propose a comparison between the tree search methods and metaheuristics on over-constrained graph coloring, in a context of minimal regulation of the parameters (chapter 5). We study the research landscape for understanding the reasons of the differences in efficiency of methods. Finally, we propose new genetic operators (crossover, mutation, diversification), the parameter setting of which is less difficult than with the classical operators (chapter 6). We conclude on the interest of investigating the neutrality networks.