In order to solve the problem that the bandpass filter parameters in resonance demodulation are difficult to select and the fault signals of rolling bearing in early failure period aredrowned in strong background noise,thefault diagnosis methodcombining variational mode decomposition (VMD)withspectral kurtosis is proposed. Firstly, the fault signals need to bereconstructed selfadaptively, so several intrinsic mode function (IMF) are obtained by VMD, and adaptive reconstruction is performed by computing the kurtosis of IMFs. Next,we can analyze the reconstructed signals by spectral kurtosis and design the bandpass filter. Finally, the working status of rolling bearingis identified through the resonance demodulation spectrum of reconstructedsignal. By processing real data,the results show that the method is more accurate than traditional resonance demodulation in diagnosing the fault of rolling bearing. Thus, it can be seen, the spectral kurtosis is reliable in selection of the filter parameters, and combining VMD and spectral kurtosis can reduce the noise and extract weak fault signal.