In order to realize signal denoising and fault feature extraction of rolling bearings,a rolling bearing fault diagnosis method based on SVD-CEEMDAN and KLD was presented.Firstly,the original signal was denoised by SVD.Secondly,complete ensemble empirical mode decomposition with adaptive noise was used to adaptively decompose non-stationary vibration signal after denoising,and the intrinsic mode functions were obtained.Thirdly,the effective intrinsic mode functions were selected for reconstruction by Kullback-Leibler divergence,then autocorrelation noise reduction for the reconstructed signal was done.Finally,the characteristic frequency of fault was shown by envelope spectrum analysis of the denoising signal.The analyzed results of bearings validate that the proposed method can be used to effectively suppress noise and the signal reflected actual fault information can be gotten.The method is practical and effective.