Abstract: Considering the instability of acoustic emission (AE) signals of different rock fracture, the method for feature extraction and comprehensive recognition of AE is put forward combining with AE parameters, Welch spectrum, EMD and BP neural network. Through the acoustic emission experiment of three different brittle rocks under uniaxial compression, stress-strain curve and AE data are obtained. Comparative analysis is carried out towards the time-frequency characteristics of AE signal of rock samples. Feature vectors, such as AE parameters, Welch spectrum, and EMD energy entropy, are integrated with BP neural network to recognize different AE signal patterns. The results show that there are similarities and differences in characteristic evolving with stress or time of AE parameters of different rocks under uniaxial compression; characteristic differences of AE spectrum and energy distribution of different rocks can be well reflected from EMD and Welch spectrum; a high recognition rate can be reached by neural network with various characteristics of different rock acoustic emission.