Early fault signal of rolling bearing usually presents nonstationary multicomponent characteristics, early fault features of bearing submerged in the background noise are difficult to identify because of the weak modulated characteristics and strong noise. Therefore, the fault diagnosis method based on variational mode decomposition (VMD) and autocorrelation analysis was proposed．At first, the noise was eliminated and the periodic components in signals were extracted by using autocorrelation analysis. Then VMD was used to decompose the denoised signal into many intrinsic mode functions and the IMFs of the biggest coefficient and kurtosis was selected and demodulated with Teager energy operator. At last, the bearing fault type was distinguished through the energy spectrum. The simulation experiments and practical engineering experiments have been carried out and the results show that this method is able to reduce the interference of noise and extract effectively the fault feature frequency, and realize accurate diagnosis for rolling bearing fault.