Abstract : Modern computing systems are no longer composed of personal computers and super computers. Many computing platforms are now available and each one has its own constraints, leading to many different objectives. Thus, the notion of performances becomes multi-dimensional. For instance, scheduling optimally (for time) and application on a computational grid is useless if no results are given when a computer crashes. Solving these problems is a present algorithmical challenge. In this manuscript, we study multi-objective scheduling through the approximation theory. We tackled four problems. The first two ones come from embedded systems while the last two ones are derived from cluster and grid computing. The first studied problem is the optimization of performances of an application running on a machine with low storage capacity. We show that using multi-objective optimization leads to solutions and informations for this problem that could not have been obtained by the classical approximation theory. The two following problems tackled the optimization of the performances of an application when computers are not entirely reliable. Different models can be studied leading to radically different optimization problems. That is why the second problem considers safety in embedded systems while the third one deals with the reliability in computational grids and clusters. The last problem is about scheduling tasks of several users on a parallel computing platform. We show how multi-objective optimization can be used to take the users' needs into account during the scheduling process.