Abstract : This work contributes to the development of Evolutionary Multi-objective Algorithms. The increasing interest for these techniques observed since the last decade is mainly due to their ability to find a (good) sampling of the whole set of the Pareto compromises in a single run unlike the traditional multiobjective optimization approaches that provide only one compromise solution which, what is more, highly depends on the subjective choice of certain parameters. Indeed, when solving the real-world multi-objective optimization problems and, in particular, design problems, it is often preferable to make the final decision from the informations as complete as possible even if an additional computation effort is needed. In this thesis, two problems from car industry are studied. The first one is the optimization of the shape of the car front end taking into account 10 objectives issued from the crash, acoustic and static mechanical domains. The second problem consists in optimizing the parameters controlling the fuel injection for the Common Rail diesel engine in order to minimize the specific consumption and combustion noise subject to the European Emission Standard constraints. An important tendency of the Evolutionary Algorithms is that these methods today "penetrate" into numerous new applicative domains in spite of the absence of a solid theoretical basis (notably, convergence proofs) as can be found for the alternative approaches. Inspired by this observation, the main motivation of this work was to contribute to the development of Evolutionary Multi-objective Algorithms in such a way as to make their application to the real-world problems most efficient. Thus, one original contribution of this thesis consists in responding to a very important lack in this domain, the lack of a stopping criterion finer then just the bounding of the number of iterations. The stopping criterion proposed in this work is intended to optimize the ratio between the solutions quality and the computation cost: indeed, in practice this is what is very often sought. Then, a new crossover operator based on the Pareto dominance relation is proposed and we show that it accelerates the progress toward the Pareto surface.