In this paper, double PCA combined with BP neural network is used. Firstly the principal component analysis is used to classify the sample data and then the BP neural network is used to recognize the unrecognized samples. The study result indicates that the new method improves the precision of recognition and reduces the training time.