In order to improve the effectiveness of bearing fault feature information extraction,the intelligent recognition of bearing fault pattern is realized,and the efficiency of fault diagnosis is improved. The method of rolling bearing fault diagnosis is proposed,which combined with de- noising of singular value decomposition and fuzzy LMD feature quantification and PNN neural network recognition. In this method,firstly,the original signal is reduced by using SVD noise reduction technique,and the LMD decomposition is used to decompose the non- stationary signal into a number of stable product function components（ PF）. Secondly,because the fuzzy entropy can be used to characterize the complexity of time series and the stability of statistical property. The fuzzy entropy of PF component is extracted to form N dimension feature vector. The PNN network model is constructed,the feature vector is input PNN model to realize the fault type identification. Finally,the performance of the PNN algorithm and BP algorithm are compared,and the superiority of the PNN algorithm is shown. The experimental data show that the proposed method has a high accuracy rate of 93. 75% in the case of a small number of data samples.