Nonstationarity is an important characteristic to represent the randomness and the variability of engineering signal, and its quantification and evaluation are the effective bases for signal analysis and pattern recognition. The typical calculation algorithms of the global non-stationary and local non-stationary are demonstrated and compared. Based on moving statistical analysis, the concept and algorithms of moving mean value, moving variance and moving variation coefficient are proposed, in order to fully characterize the varying details in time domain and the diversity of different signals. By the nonstationarity analysis of different mechanical fault signals and electrocardiogram signals, the evaluation capability and applicability of different algorithms are analyzed and compared. The results indicate that the moving statistics method has outstanding theoretical foundation, precision and stability in the evaluation method of nonstationarity.