According to the problems that the fault features identification of rolling bearing vibration signal, a method for fault feature extraction was proposed base on the improved EEMD with multiple population differential evolution (MPDE) and adaptive resonance demodulation technique (ARDT). Firstly, in order to solve the problem that the EEMD􀆳s parameters selection depending on individuals’ experiences, an evaluation function for distribution characteristics of extreme value points was established. It was used to optimize white noise amplitude using MPDE. Then, EEMD adaptive decomposition was implemented. Secondly, effective signals of the decomposed IMF components were reconstructed using criteria for kurtosis and relativity. The signal de-noising process was realized. Finally, the center frequency and bandwidth of band-pass filter was adaptively determined based on ARDT, and the fault characteristic frequency was extracted using envelop demodulation analysis. A simulation signals and a rolling bearing test results show the validity of the proposed method.