Aiming at the disadvantage of wind power bearing fault vibration signal, such as low frequency, not smooth, nonlinear, difficult to detect, it puts forward a new fault diagnosis method for bearing. First of all, the Ensemble Empirical Mode Decomposition (EEMD) noise reduction method was proposed based on integrated correlation index and kurtosis to make noise reduction of acceleration signal of bearing vibration, then the combination of EEMD and cepstrum was used process the vibration signal, which could not only improve the signal’s noise ratio, but also inhibit the mode mixing phenomenon in the EMD decomposition process. It improves the accuracy of fault diagnosis and fully shows the feasibility of its application in wind turbine bearing fault diagnosis system.