In order to extracting the characteristic frequency of rolling bearing signal, a feature extraction method based on adaptive partly-ensemble local characteristic-scale decomposition (APLCD) and wavelet package transform (WPT) is proposed, or APLCD-WPT for short. First, APLCD is employed to process vibration signals of rolling bearings, and it can add noise to improve the signal extreme-point distribution in extracting intrinsic mode component by changing the frequency variation. Then, WPT is used to trim less modal mixing problem, which can extract characteristic frequency signal of rolling bearing. Finally, the vibration signal of horizontal spiral centrifuge is analysed based on this method. The results show that APLCD-WPT can effectively suppress the mode mixing to accurately extract the characteristic frequency signal.