In the light of the difficulty of rolling element bearing defect diagnosis under variable frequency while using traditional spectrum analysis， a new method based on fast Hoyergram and improved variational nonlinear chirp mode decomposition （VNCMD） is proposed. First， fast Hoyergram is used to determine the resonance frequency band where the bearing fault impact locates. Then bandpass filtering is utilized to extract the component of rolling bearing vibration signal and the result is mixed with the signal after lowpass filtering. Secondly， the ridges of rotating frequency and bearing fault frequency are extracted based on multi-component collaborative speed estimation， and used as the input parameter of VNCMD to extract the rotation and bearing fault impact components. Finally， the type of rolling bearing fault can be determined by characteristic frequency ratio. Compared with ensemble empirical mode decomposition （EEMD）， the proposed method can extract more accurate time-frequency ridges and obtain correct components through signal decomposition. Both the simulation and experimental results demonstrate the effectiveness of the proposed method.