In this paper, a signal processing method of combining the empirical mode decomposition with the vibration mode analysis is proposed. The empirical mode decomposition is applied to filter out the high-frequency modes of a bearing vibration, then, the residue signal is analyzed by the vibration mode analysis to derive the envelope signals of the bearing vibration in the low-frequency modes. Accordingly, the envelope spectra are applied in the defect diagnosis of the bearing vibration. First, in the empirical mode decomposition a fourth order polynomial function is applied to estimating envelopes for the extreme locations of the bearing vibration, and to deriving the first intrinsic mode function which would contain two high-frequency modes. Secondly, the vibration mode analysis is applied to estimating two envelope signals of the low-frequency modes for the residual signal which is derived from the bearing vibration with filtering out the first intrinsic mode function. Finally, by estimating the vibration spectra for the envelope signals the running condition of the bearing could be diagnosed. From the simulated and empirical studies, the algorithm of combining the empirical mode decomposition with the vibration mode analysis is proved to be effective in the defect diagnosis of bearing vibrations.