In connection with the processing characteristics of aviation parts, acoustic emission(AE) signals of tool which in different wear state were acquired. Time-frequency analysis and wavelet transform were utilized on the AE signals. Fast fourier transform and db8 wavelet decomposition were used to extract the amplitude root-mean-square value and the main band energy value which were considered as eigenvectors of AE signals. Then the eigenvectors were normalized and taken as input vector for the training of wavelet neural network. Parameter adjustment algorithm was applied to add momentum term to weights and threshold for amendment in the wavelet neural network. The results indicate that the frequency range of tool AE signals which sensitive to different wear states is between 10-150�CkHz. The error between the trained wavelet neural network actual output and expect output is less than 0.03. Different tool wear states can be recognized correctly and effectively by this method.