Outliers as well as outlier patches, which widely emerge in dynamic process sampling data series, have strong bad influence on signal processing. In this paper, a series of recursive outlier-tolerant fitting algorithms are built to fit reliably the trajectories of a non-stationary sampling process when there are some outliers arising from output components of the process. Based on the recursive outlier-tolerant fitting algorithms stated above, a series of practical programs are given to online detect outliers in dynamic process and to identify magnitudes of these outliers as well as outlier patches. Simulation results show that these new methods are efficient.