Abstract: Aiming at the composite fault diagnosis of the rotor failure and weak roller bearing fault, a fault diagnosis method was proposed based on the wavelet transform and the ensemble empirical mode decomposition by separating the coupling features of the composite fault and extracting the frequency of fault signals. Firstly original signals were decomposed and reconstructed by using the wavelet. Then non-modulation low-frequency fault feature was extracted using FFT for the low-frequency signals obtained from the decomposition and reconstruction of original signals. The high-frequency modulated signals from the decomposition and reconstruction of original were analyzed by envelop demodulation based on the ensemble empirical mode decomposition （EEMD）, and the modulated fault feature was extracted. Analyzing results of engineering signals indicated that the method can extract the composite fault feature of rollin bearings successfully.