Based on Mach-Zehnder interferometer, the fiber-optic intrusion recognition system was established to realize high precision recognition system. By introducing the principle of multiple regression statistics, the multiple regression equations between signal characteristics parameters and dependent variable were set up. Optimal selection of characteristics was realized by the administration of a significance test on the regression model with method of gradually introducing. Classifier was designed based on perception with optimizing characteristics to establish fiber-optic intrusion recognition system. The test results indicate that the recognition system based on multiple regression has accuracy of identification is 83.33%.