This paper describes a new probabilistic fusion methodology based on Shannon's entropy, whose goal is to reduce the combination space by explicitly representing the notions of source redundancy and source complementarity in form of entropy measures. This fusion methodology called Entropy Fusion Model (EFM) is defined and implemented in three steps: modeling step, combination step and decision step. The EFM approach shows how an information fusion problem can be formulated by using entropy criteria minimization as a basis for guiding the fusion system to the best fused information. The main advantage of such a fusion approach is to optimize the choice of measurements provided by information sources in order to improve the performance of the information fusion system. Experimental results from an application to mobile robotics are presented illustrating the performances and the robustness of the Entropy Adaptative Aggregation (EA2) resulting algorithm.