Using the multilayer forward neural networks(MFNN) based on back propagation(BP) algorithm, the relationship between the process measurements and fault type was constructed, the identification of the normal state and fault state was achieved. For improving the accuracy of identification, wavelet technology was used, the activation function of MFNN was modified and wavelet neural network (WNN) was constructed. The simulation results of a classical reaction process of chemical reactor show that WNN has higher accuracy than MFNN for fault identification.