Aiming at the strong background noise involved in the signals of rolling bearing and the difficulty to extract fault feature in practice, a new fault diagnosis method is proposed based on Dual-tree Complex Wavelet Transform (DT-CWT) and AR power spectrum. Firstly, the non-stationary and complex vibration signal is decomposed into several different frequency band components through dual-tree complex wavelet decomposition; Secondly, Hilbert envelope is formed from the components which contains the fault information. Finally, the auto-power spectrum can be obtained by auto-regressive (AR) spectrum. The noise interference was eliminated effectively, and the effective signal information was retained at the same time. Thus, the fault feature information was extracted. In this paper, the fault test and the engineering practical fault data of rolling bearing were analyzed by dual-tree complex wavelet transform and AR power spectrum. The results show that the noise of the vibration signal was eliminated effectively, and the fault feature were extracted. The feasibility and effectiveness of the method were verified.