Vibration signals of faulty bearings are usually nonlinear and nonstationary under heavy background noise.As a result

the feature extraction of the vibration signals is always a quite difficult problem.To improve fault diagnosis results for bearings

the fault diagnosis method combination of empirical mode decomposition（EMD） and spectrum kurtosis（SK） is proposed.For one thing

the EMD method is utilized to filter the trend and noise components from the faulty signal in term of the statistical characteristics of the normalized white noise.For another

the bandwidth and central frequency of a bandpass filter which is going to be exploited to filter the purified signal by the EMD method are estimated by using SK.Moreover

according to the envelope spectrum of the signal filtered by the above bandpass filter

the bearing fault is identified.The results show that the SK method can be effectively improved by the EMD filter and a feasible method for the fault diagnosis of bearing is provided.