This thesis is part of an international research program (IRG Cognac) initiated by the engine manufacturer SNECMA (SAFRAN group) involving ENS Cachan, UnB, ENSMA, CNRS, Snecma, Turbomeca et Messier Bugatti Dowty. The thesis focuses on the effect of a stress gradient in fretting fatigue. Fretting-fatigue refers to the damage process localized at the frontier of the contact between two contacting bodies subjected to fatigue loadings. The prediction of this phenomenon is of major importance in determining, for instance, the lifetime of fan's disc. In the vicinity of the contact front, the stress field inherited from the contact loads is maximal at the surface and displays a strong gradient from the surface. It was shown in this thesis, for a TA64V alloy, that local approaches, based on local stresses at the most critical point, are not appropriate to predict fretting fatigue lives. As a matter of fact, short cracks initiated at the most critical point may stop if the stress decay from the surface is strong enough or may continue their growth, up to the failure of the component, if the stress gradient from the surface is not string enough. A second difficulty is the multiaxial and non-proportional nature of the loading conditions. Fatigue-fretting stems from the combination of loads that have neither the same spatial distribution nor the same time-dependency. In fretting-fatigue tests, three loading components are considered, the fatigue loading of the component (cyclic), the normal part (assumed to be constant) and the in-plane part (cyclic) of the loads between the two contacting components. To quantify the effect of the stress gradient, tests were carried out on a fatigue testing contact bench developed at the University of Brasilia, with experimental conditions ensuring different stress gradient while keeping the maximal stress the same. Damage mechanisms were studied using post-mortem analysis and optical microscopy on the contact elements tested. The prediction of the fretting fatigue life was done using different approaches. The first one is based on the Critical Distance Method and a fatigue criterion. The second is based on a K-based short crack arrest method. Finally, a new criterion was proposed. This method considers a generalized von Mises yield criterion for the crack tip region and accounts for the T-stresses in the asymptotic LEFM development. 4 Remerciements -Agradecimentos -Acknowledgements Avant tout autre chose, je souhaiterais remercier Jean Mériaux, Sylvie Pommier et Alex Araújo qui ont suivi et dirigé ma thèse. Merci pour leur confiance, leur patience et leur ténacité. Au delà de leurs compétences scientifiques reconnues. Je remercie l'ANRT et SNECMA (Groupe SAFRAN) pour le cofinancement de ma thèse. Et je remercieégalement Fabienne Fong pour son soutien et son aide. Me arriscarei no português para agradecer as pessoas que marcaram a minha estadia no Brasil. Obrigado a Jordana Peixoto, Julia Ayla, Patricia Vitorino, Elissa Massote, Guillaume Tessier, Wallison Martins e todos os outros. Graças a...