According to the abnormal data partitioning of the statistic diagnosis and the causes of the error data in dam monitoring, the abnormal data observed can be categorized into random error, gross error, system error. The influential data is distinguished and then the processing methods for different errors are studied using the mean shift model, including the t-test method for the gross error detection based on the model disturbed values and the Cook-distance-test method for the influential data based on the influence of the model disturbance on the model fitting parameters. Taking typical original monitoring data of dam displacement as an example, rationality test was performed using the proposed statistical diagnosis methods. The results indicate that these methods can effectively distinguish gross errors and influential data, with which the accuracy of further model analysis can be significantly improved.