Feature extraction of rolling element bearing’s compound faults is a challenging task due to the complexity and the mutual coupling phenomenon among the kinds of faults. A new method based on cyclic wiener filter with constructed reference signals is proposed in the paper. The reference signals of the rolling element bearing’ inner race fault, outer race fault and rolling element fault are created respectively based on the rolling element bearing’ theoretical fault frequencies. Here, the created signals are used as the expected responses. Then the observed compound faults signal and the constructed reference signal are input into the cyclic wiener filter together. At last, the envelope demodulation method is applied on the filtered signals respectively and satisfactory fault feature extraction results are obtained. The effectiveness of the proposed method is verified through simulation. Furthermore, the advantages of the proposed method over other signal handling method such as spectral kurtosis (SK) are verified through experiment.