Aimed at the problems of rolling bearing vibration signal contained a lot of noise and presented the characteristics of nonlinear, nonstationary and etc., which brought a great difficulty for accurate fault feature extration, a novel rolling bearing fault feature extraction method based on singular values difference spectrum denoising combined with empirical mode decomposition (EMD) was put forward. In the first place, the signal was reconstructed in phase space and then the singular values difference spectrum theory was used to realize denoising. After that, several intrinsic mode components (IMF) containing local signal characteristics were obtained by performing EMD on the denoised signal. Finally, performed Hilbert envelope demodulation on the largest correlated IMF and original signal, fault feature characteristic frequency was extracted. The experimental result shows that the method can not only effectively denoising the signal, but also accurately extract the fault feature information of rolling bearing.