According to the frequency domain energy distribution differences of bearing vibration signal in the different failure modes, rolling bearing fault pattern recognition technology based on the orthogonal wavelet packet decomposition and BP neural network is proposed. The orthogonal three layer wavelet packet decomposition for rolling bearing vibration signal is carried out to get the third layer wavelet packet decomposition coefficients from low frequency to high frequency，then the different frequency band signal are reconstructed respectively to extract energy features by means of wavelet packet decomposition coefficients. Using the energy feature vector of different frequency band as the model input of the BP neural network model，a large number of samples are trained to get the network pattern recognition model for different bearing fault，then use several groups of test data are used to verify the BP network models to discrimination the type of rolling bearings fault. The test results proved that the method integrated the Wavelet packet decomposition with BP neural network can identify the fault of rolling bearings more accurately.