At present, the dynamic analysis for non-stationary stochastic noise of ceramic paste inner stress is a difficult problem because a non-stationary stochastic noise has a time variant model and dynamic characteristics. In this paper, dynamic analysis ability of Short Time Power Spectrum (STPS) was systematically studied and improved in accordance with theoretical research and simulation analysis. The pressure signals between ceramic paste and head inwall of vacuum pug mill were studied so as to obtain the distribution of the ceramic paste inner stress. Experiments about the pressure signals had been carried out to verify the research result. Theoretical analysis and experimental result shows that: STPS with filtering performs well in analyzing non-stationary stochastic noise of ceramic paste inner stress in mutation situation.