Abstract: Wavelet packet,which has the function of analyzing non-stationary signal locally, can solve the problem of poor resolution in high frequency part with wavelet analysis. Hereby, a method of rolling bearings fault diagnosis based on wavelet packet energy spectrum was proposed. Firstly, the vibration signals were decomposed into individual frequency bands by wavelet packet, and energy ratio of each band was obtained. Then, the energy spectrum scale between the normal vibration signal and fault vibration signal was compared to identify the fault frequency band. The fault signal frequency band were reconstructed, the reconstructed signals envelope were demodulated with Hilbert transform, and the fault frequency was extracted. Experimental results verify the feasibility of the wavelet packet energy spectrum for fault detection of rolling bearing.