In order to distinguish and analyze the vibration signals of similar working states effectively, in this paper, the method of differential bispectrum based on time series was proposed to amplify the difference of similar signals. Milling machine vibration signals which obey a stationary random process were set to the research object and their bispectra were highly similar and difficult to distinguish. In this paper, the comparative analysis result of bispectra and bispectral diagonal slices after multiple differential demonstrates that the spectra which have a high degree of similarity before differentiated become different after differentiated, and some of these differences increase with the differential times. So the analysis indicates that the differential method is an effective feature extraction method, and with this method the vibration signals of highly similarity can be distinguished and analyzed. The coupling components in signals can be enhanced. Because in various machining systems, many parameters such as speed, material and roughness are relative to bispectral features, so the parameters can be adjusted according to the analysis results to achieve the purpose of improving system functionality.