For the composite fault of wind turbine rolling bearing in actual working conditions， due to the interaction between multiple faults which interfere with each other， making the composite fault feature difficult to separate. A method for separating and extracting composite fault characteristics of wind turbine rolling bearings is proposed based on spectral kurtosis （SK） and multipoint optimal minimum entropy deconvolution adjusted （MOMEDA）. Firstly， the spectral kurtosis analysis is performed on the composite fault signal， and the resonant frequency band with larger energy is selected. The band-pass filter is constructed to filter the corresponding resonant frequency band， and the envelope signal is analyzed by the envelope spectrum to separate the single fault feature. Then， the multipoint kurtosis spectrum analysis is performed on the filtered signal that fails to realize single fault feature extraction， and the fault period is determined. The subsequent separation and extraction process is completed by using MOMEDA. The simulation signal and engineering application analysis results show that the method can effectively and accurately realize the separation and extraction of bearing composite fault features.