Rolling bearings are one of the most widely used and most easily damaged components in mechanical equipment. Extracting the vibration signal of the rolling bearing can give us a better grasp of the equipment’s operational state. In practical applications, traditional wavelet package transform (WPT) due to a defect itself MALLAT algorithm cannot accurately extract the characteristic frequency of the signal. Complementary ensemble empirical mode decomposition (CEEMD) can effectively restrain the mode mixing problem, but cannot completely avoid it. In order to accurately diagnose rolling bearing defects, we propose the WPT-CEMMD feature extraction method, based on CEEMD and WPT. Combining the the two methods could not only effectively solve the problem of mode mixing after CEEMD decomposition, but also eliminate the influence of the spurious frequency component and frequency aliasing after WPT treatment. Both simulations and a case of the working frequency of extraction demonstrated the efficacy of the proposed method.