Abstract: In order to estimate traffic states more accurately by fusing multisensors in intelligent traffic surveillance system, this paper presented a federated evidential fusion method, in which the evidence theory and the federated filter are integrated. This method is successfully applied to a real urban traffic network for realtime traffic state estimation. Also it was testified that this approach can not only overcome the drawback that the evidence theory can not deal with conflict exactly, but also enhance the realtime performance and robustness of the evidential fusion system, because the structure of the federated filter makes it possible to combine the temporal information and the reliability of sensors into the fusion system.