The vibration signals of a rolling bearing usually exhibit characteristics of multi-carrier and multi-modulation. In light of this problem, an eigenvector extracting method based on the local mean decomposition (LMD) energy feature is proposed. The method is combined with the support vector machine and applied to fault diagnosis of rolling bearings. The complicated modulation signal is decomposed into a set of product functions (PF) using LMD. Then, the energy moment of PFs containing the fault characteristics is obtained by calculating the integral of PF as input parameters of a support vector machine classifier to sort the fault patterns and degrees of rolling bearings. To analyze the vibration signals acquired from the bearings with inner-race, outer-race or element faults, experimental results indicate that the proposed method can effectively extract the fault characteristics, accurately identify the type of rolling bearings and achieve a comparatively high correction-identification ratio of the fault level.