Aiming at the problem that it is difficult to obtain the fault information of rolling bearing vibration signal accurately, a new bearing fault feature extraction method based on multilayer noise reduction was proposed. The original bearing signal was processed according to the principle of wavelet packet transform to eliminate the noise interference; several IMF components were obtained by processing the transformed vibration signal with empirical mode decomposition(EMD), and the crosscorrelation coefficients between the obtained components and the transformed signals were calculated, and the useful components were selected according to the correlation coefficient criterion to complete the reconstruction of the vibration signal. Then, by autocorrelation processing, aliasing interference signals in EMD decomposition and reconstruction signals were eliminated to realize the extraction of bearing characteristic signals. Finally, the reconstructed signal after denoising was demodulated, and the signal envelope spectrum was obtained and analyzed to obtain the required fault features. The vibration test results shows that using the method, interference and noise in original signal can be eliminated effectively, clear fault vibration signal can be separated and useful fault features can be obtained.