Abstract: Fast and accurate power system disturbance detection can provide effective guidance information for subsequent disturbance analysis, and the wide area measurement system (WAMS) is widely used to provide a powerful data base for disturbance detection. Based on PMU measurement data, this paper proposes a disturbance event detection method considering PMU bad data. First, the behavioral characteristics of PMU abnormal data are analyzed to reveal the differential characteristics of disturbance events and PMU bad data. Furthermore, a PMU abnormal data initial screening method based on the combination of differential Teager-Kaiser energy operator and 3Sigma criterion is proposed to avoid the problems of low intensity disturbance miss detection and repeated detection of disturbances. Then, the dynamic time warping and the maximal information coefficient are used to calculate the spatio-temporal similarity among different PMUs and the correlation among different measurements within the same PMU, respectively. And it is used as features to characterize the differences of disturbance events and PMU bad data. Finally, the obtained comprehensive metrics are analyzed by a local outlier probability algorithm to achieve accurate detection of disturbance events in scenarios containing PMU bad data. Based on the IEEE 39 system, the actual grid model and the filed PMU data, the proposed method is verified to have good accuracy, real-time and generalization capability.