The working environment of rolling bearing is usually seriously disturbed by noise, so it is difficult to detect its fault. To solve this problem, a method of optimizing VMD parameters based on improved firefly algorithm was proposed. The signal was analyzed by fast spectral kurtogram to get the best parameters of the bandpass filter, then the signal was filtered. The optimal parameters K and α of VMD were obtained by the optimization of firefly algorithm. By using the parameters obtained, the signal was decomposed into several IMF components and then it was reconstructed based on correlation coefficient and dispersion entropy. Finally, the reconstructed signal was demodulated by the Hilbert envelope to get the fault characteristics. Through the analysis of experimental data and the comparison with LMD decomposition, this method can be used to detect bearing fault features reliably.