Aiming at the problem that the conventional time-frequency domain feature extraction method is limited due to the non-linear, non-stationary and strong noise characteristics of the rolling bearing fault signal, a bearing fault feature extraction method based on enhanced combination difference multiply morphological filter is proposed. Based on the analysis of the positive and negative shock pulse extraction characteristics of the four basic operations of mathematical morphology, the cascade, difference, and multiply are constructed to construct a new combination difference multiply operator (CDMO) with the ability to simultaneous extraction of positive and negative shock pulses, and take advantage of the gradient multiply operation that is more sensitive to pulse extraction, to achieve sufficient extraction of fault information. Introduce the fault characteristic frequency ratio index to optimize the parameters of the CDMO structural elements, modify the geometric characteristics of the signal to be processed, and extract the signal characteristic information that matches the structural elements. On the basis of CDMO filtering, with the help of third-order cumulant slice spectrum technology which can suppress gaussian noise and highlight the advantages of secondary coupling components, accurately extract fault feature frequencies and their multiplications, enhance the ability of bearing fault feature extraction and suppress noise interference. Relying on actual engineering signals from two different sources and comparing with classic fault feature extraction methods, the effectiveness of the proposed method is verified.