The signal acquired by acoustic emission method to monitoring the complex mechanical seals usually has a low SNR signal, which makes the classification of the working condition of mechanical seals difficult. A new method was presented to classify the condition of mechanical seals based on acoustic emission and wavelet neural network. This method was combined of the wavelet and neural network, and extracted characteristics based on time domain and wavelet pack analysis to make full use of the useful information in acoustic emission signal, which could represent the working condition of mechanical seals well. With an experiment on the dynamic seals for rotating shaft, the method mentioned above was used to monitor the working condition. The experiment proves that this method can classify the working condition and fault pattern of complex mechanical seals effectively.