According to the shortcomings of traditional dual-tree complex wavelet transform which has spectral aliasing, a novel approach to multi-faults diagnos is of bearing based on improved dual-tree complex wavelet transform is presented. Firstly, the bearing fault vibration signals are decomposed into various frequency band signals. Then the band signals are decomposed using empirical mode decomposition technique, and the maximum energy component is found. In the end, the envelope spectrum of maximum energy components is computed. Therefore, the characteristics of the bearing faults can be recognized according to the envelope spectrum. The simulative and experimental results show that not only the spectral aliasing is eliminated, but also the frequency band selection accuracy and signal noise ratio are improved, the multi-faults of the bearing can be effectively detected.