Abstract: After a brief review of the methods of detecting, identifying and controlling outliers, a new outlier snooping method is developed for the time series of satellite gravity data based on least square collocation. The main idea is to predict observable of the possible outlying point from neighboring data and to compare the prediction error with its estimated variance, and then perform outlier snooping according to a certain test statistic and a given significance level. Based on the above idea, the test statistic of t-distribution is constructed and a new statistical test of moving window is proposed. A simple correlation function, the determination of window width and a detail iteration testing procedure are presented. A numerical example from the simulated Tzz of GOCE satellite is performed. The results show that the new testing method can efficiently detect and identify the outliers.