In view of the fact that the impact signal of rolling bearing is easily submerged by noise and nonstationary characteristics, and the problem that the effective information in the IMFs cannot be fully utilized when the traditional CEEMDAN is used, a rolling bearing fault feature extraction method based on weighted average complete ensemble empirical mode decomposition with adaptive noise (WACEEMDAN) and modulation signal bispectrum (MSB) for rolling bearings is proposed. First, the CEEMDAN is used to decompose the collected nonstationary vibration signals into several inherent modal functions with stationary characteristics IMFs. Then, a new type of index is constructed, which emphasizes sensitive component: correlationkurtosis value, the index is used to weight each IMFs and reconstruct it into WACEEMDAN signal. Finally, the MSB is used to decompose the modulation components in the WACEEMDAN signal and extract the fault characteristic frequency. The results show that: by using the general bearing data set of Xi’an Jiaotong University and our test bench, the proposed WACEEMDANMSB method can accurately extract the characteristic frequency of bearing faults, thus, verify the effectiveness of the WACEEMDANMSB method.