Abstract: In view of the problem that diagnosis effect of traditional fault diagnosis method of bearing of asynchronous motor is bad for local defect and initial fault of bearing, the paper proposed a method of using wavelet-packet theory and EMD to extract fault characteristic frequency of bearing. The method uses wavelet-packet theory to delete noise and differentiate frequency band for original signal at first, then uses EMD to decompose the signal got by wavelet decomposition and reconstruction so as to get IMF, at last the spectrogram of each IMF is obtained through time-frequency transform and diagnosis result can be gotten according to fault characteristic frequency. The experiment result proved that the method can extract fault characteristic frequency effectively and easy to judge fault type.