The reliability of safety parts is a major issue for an automotive manufacturer. Fatigue failure is, however, a phenomenon that is difficult to analyze because it depends on the one hand on the manufacturing of the part, its geometry and the mechanical properties of the various materials used, and on the other hand on the external loads it is subjected to. In order to better design safety parts against fatigue failure, numerical simulations and tests on real size prototypes are carried out by the design office. Deterministic fatigue criteria are then used to identify the critical zones of the part. If these criteria prove to be effective on experimental test data with standardized specimens, they are less effective for bench tests with prototypes: the variability inherent to tests on prototypes is poorly addressed by deterministic criteria, not to mention the errors due to numerical simulations. We then propose to use statistical methods on the one hand to improve the deterministic criteria; on the other hand to build new fatigue criteria.