Dynamic assessment of structures under ambient loads is an important applied field of the structural mechanics and synamic system theory, which is the subject of many researchs for about fifteen years. Registered within the framework of the French national project Evaluation dynamique des Ponts controlled by the Laboratoire central des Ponts et Chaussées (LCPC), the work presented in this thesis is a contribution to these researchs. It concerns the study of two tools for their relevance in the field of modal identification of structures under random dynamic excitations : subspace methods and random decrement method. A broad place is devoted to the subspace methods which are the subject of a detailed theoretical and algorithmic analysis intended for determining their possibilities and to clarify some ambiguities. The main contribution of this work concerns the random decrement method. We give first an original presentation adapted to discrete time context. Then, using recent results, we justify rigorously asymptotic results of practical and especially statistical interest. Then we show the reason why this method, whose usual theoretical justification rests on an assumption of stationary of the excitation, still works when this excitation is an impulse or a batch of impulses. The practical interest of these methods in the field of modal identification is tested to two applications: the first one concerns a beam tested in laboratory, the second one is a railway bridge tested in situ.
