It can be easily understood that carrying out a complete optimization (whether multidisciplinary or not) with a highly refined model can rapidly lead to prohibitive computing costs. Conversely, if a relatively coarse model is used, an optimum can still be obtained, but with little confidence in the results. Thus, it appears to necessary to be able to vary the degree of refinement of a model throughout the optimization process in order to make this process both accurate and affordable in terms of computing cost. However, such a multilevel model optimization process does raise numerous questions. we will try to answer these questions and obtain a global picture of the main multilevel model optimization approaches available. This will enable us to propose original and innovative multilevel methods.