To solve the problem that the wavelet singular spectrum entropy can not effectively discriminate the signals of inner and balling faults,this paper attimpts to diangnose bearing faults by using the maximum wavelet singular spectrum of the bearing vibration signals.The maximum singular value of the wavelet matrix resulting from the multi-resolution analysis of the bearing vibration signal was used as the classification feature.A bearing fault diagnosis experiment was conducted and its results were compared with that of the wavelet singular-spectrum-entropy method.The effects of the wavelet base selection,window width on the proposed method were analyzed.The results show that the method is highly capable of bearing fault diagnosis and has better performances of stability and robustness.