In the light of the identification of the faults type of rolling bearing, which is hard due to the non-linear and nnon-lineary characteristics of the fault signals, a method of fault identification is proposed. It consists of the experience wavelet transform (EWT), multi-scale permutation entropy (MPE) and GG (Gath-Geva) clustering algorithm. First of all, the original signals of rolling bearing are decomposed into many intrinsic mode components based on the EWT decomposition. Then, the state features of the rolling bearing are preliminary extracted; the optimal modal component is selected with correlation analysis, and the permutation entropy is calculated in multiple scales. Finally, the principal component analysis (PCA) is used to reduce the dimension of the entropy feature vector for visualization, and low features subset is introduced into the GG clustering algorithm to realize the fault diagnosis of the rolling bearing. Comparisons with other mode combination method show that the proposed fault diagnosis method has certainly advantages，which better fault recognition effect.