To resolve the security problem of SIP, a detection method based on statistics is proposed. This method fuses multiple attributes of user's behavior into modules based on D-S evidence theory, and by such fusion it achieves the complementation of modules and the detection of multiple attacks. It also differentially considers the possibility of attack and legality of user by classifying domains,which improves accuracy of detection. Experiment and analysis results showed the high accuracy of the proposed method, by which the SPIT can be detected efficiently.