Abstract: The environment radiation monitoring system in nuclear power plant continuously monitors γ dose rate. If the dose rate exceeds a predetermined threshold, the system will issue a warning. However, the radon and thorium daughters effects and the system noise and their periodic changes will result in fake alarm. In order to study these factors, a support vector machine model was established to classify the reasons of threshold exceeding. The model was tested by more than 2 000 groups of historical data. The results show that it can accurately classify the reasons, and the accuracy rate is more than 98%.