Aiming at the problem that small fault signal is easy to be drowned by noise in strong noise,a bearing fault diagnosis method based on maximum second-order cyclostationarity blind deconvolution（CYCBD）and complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN）was proposed.The fault frequency of vibration signal was calculated according to the fault frequency formula,the corresponding cycle frequency set was set according to the fault frequency,and CYCBD was used to filter the original signal to highlight the periodic pulse component in the signal which was concealed by noise,so the effect of improving SNR was achieved;the CEEMDAN processing was performed on the processed signals to obtain a series of IMF components,then several components with high kurtosis value were selected from IMF for reconstruction;the fault frequency was extracted from Hilbert envelope spectrum of the reconstructed signal.The effectiveness of the proposed method was verified through simulation and experimental research by using the simulation signal and the bearing data set from Western Reserve University.