In order to extract pathological features of heart sound signal accurately, an algorithm for extracting the sub-band energy is developed based on the wavelet packet. The db6 wavelet is taken as the mother function, and the best wavelet packet basis of heart sound signal is picked out. Then, various heart sound signals are decomposed into four levels and the wavelet packet coefficients of the best basis are obtained. According to the equal-value relation between wavelet packet coefficients and signal energy in time domain, the normalized sub-band energy of the best basis is extracted as the feature vector. Based on BP network, seven identification models for seven kinds of heart sound were trained separately. Then, these models were tested by using 70 heart sounds, and the mean of identification accuracy is 72.9%.