According to the problems of sifting stopping criterion and selection of extremums in empirical mode decomoposition， a sifting stopping criterion based on energy ratio is defined, extrema locations and extremums are determined by adopting parabolic interpolation fitting, and the decomposition results of simulated signal show the superiority of the proposed method. Finally，the improved empirical mode decomposition and Hilbert spectral are combined and applied to rolling bearing fault diagnosis. The experiment result shows the effectiveness of the proposed means.