Abstract: In this paper,a novel robust BP neural network based on the maximum likelihood method is presented.The error function of the proposed algorithm is constructed with the maximum likelihood method.Compared with the classical BP algorithm,the new robust BP algorithm not only considers the influence of noise to network learning but also may learn the system parameters from the global viewpoint,so the robustness of the proposed algorithm can be enhanced.The experimental results demonstrate the feasibility and the superiority of our approach here.