The multi-sensor fusion approach for multi-vehicle localization presented in this paper is based on the use of Bayesian network in order to fuse measurements sensors. For each vehicle, a Bayesian network is implemented to fuse measurement of embedded sensors. For the train of vehicle localization, a global Bayesian network is implemented in which we have modelled vehicles interconnections. The leader vehicle is supposed to be equipped by especially accurate sensors. With this approach, one can see that the follower's geo-positions computing are quite improved in using the Leader vehicle path and followers relative positioning provide for each follower using a rangefinder. Real data sensors are used to validate and to test the proposed approach. Experimental results are presented to shown approach performance.