The BP neural network is studied and applied to tone recognition in this paper,which is a frequently used pattern classifier.Based on analysis of training and recognition processes of BP neural network,we extract feature data reflecting the tone characteristics to build the classification eigenvector.The classifier is designed as a 3-layer BP neural network with one hidden layer,and applied to recognize the four tones of Mandarin Chinese.The recognition results by BP neural network with different hidden layer nodes are compared.The experiment result indicates that the audio feature extracted by this classifier is valid for Mandarin Chinese,and verifies its good classification performance.