The problem of intelligent fault diagnosis for a class of affine nonlinear system with incompletely observable
states is considered. Firstly, the design problem of the observer is proposed by means of diffeomorphism, and then the method of fault diagnosis based on the fault function provided by RBF neural network for the system is given. The design observer not only guarantees the observer stable but also identifies the occurrence of the fault by obserbering error signal, which guarantees the robustness of fault diagnosis algorithm and the stability of fault system. At the same time, the method is helpful to design the normal high gain observer. A simulation experiment shows the effectiveness of the proposed approach.