Using acoustic emission sensor to acquire the tool cutting signal, a method is proposed to discern the degree of tool wear by means of Back Propagation (BP) neural network. With this method, the original acoustic emission signal sample that after high-pass filtering was input into BP neural network directly to training, relying on the nonlinear mapping ability of neural network to make the neural network classified the signals that produced by different wear degree of cutting tools. What's more, which type of the unknown signal belonging to could be verified accurately. Compared with the traditional method, with this method, the link of extracting characteristic value artificially which is waste of time and energy was left out. The influence of the number of neurons to the neural network training and recognition was researched, and the identification precision of the neural network was improved. The experimental results show that this method can predict tool wear degree accurately.