In recent years, there were been increasing researches focusing on the application of artificial neural networks in structural damage identification. Most of them perform well with numerical examples under error-free conditions, but become worse when the experimental data are polluted with measurement noise. In this paper, a dynamic approach based on PNN for damage identification of composite materials was proposed. By using wavelet series, the features of signals were extracted and input to PNN for training the network and identifying the damages. A performance comparison between the PNN and BPNN for structural damage identification was carried out. The results show that the proposed method can more exactly identify the faults than the BP neural network.