User load data monitoring is the basis of demand side management. Non-Intrusive Load Monitoring (NILM) is an important development direction of load monitoring, and event detection is a key point in NILM. In this paper, the Bayesian information criterion which is suitable for the model selection problem is modeled and applied to event detection for the first time. Fast event detection algorithm is used to reduce the false alarm rate of event detection algorithm based on Bayesian information criterion, which can solve the problem of missing detection point in CUSUM algorithm. Finally, a real data set is used for testing. The experimental results show that compared with CUSUM algorithm, the event detection algorithm based on Bayesian information criterion can achieve better detection accuracy and can significantly improve the speed of computation. This work is supported by National Natural Science Foundation of China (No. 51437006) and Ministry of Higher Education and Scientific Research of Tunisia.