Aiming at the early fault feature information of rolling bearings is difficult to identify, and the parameter setting of band-pass filter depends on the user experience, which makes the resonance frequency band not be effectively determined and extracted, the concept of amplitude entropy of frequency band is proposed. On this basis, the dual-tree complex wavelet transform and Teager energy spectrum are combined, and a rolling bearing early fault feature extraction method is proposed based on dual-tree complex wavelet transform adaptive Teager energy spectrum. Firstly, original fault signals are decomposed into several different frequency components through dual-tree complex wavelet decomposition, and the frequency amplitude entropy of each sub-band is calculated. Then the entropy are arranged in ascending order and in turn as a threshold to extract the entropy value greater than the threshold value of the sub bands. The optimal threshold is determined based on the kurtosis index, thus the resonance band is extracted adaptively and effectively. Finally, the fault characteristic frequency of the bearing could be accurately identified from the energy spectrum of the resonance band. The signal simulation and experimental data analyses verify the effectiveness of the proposed method.