As the collection of moving object data become much easier, event-based outlier detection such as congestion in trajectory data are becoming increasingly attractive to data mining community. Most of the existing methods only perform the trajectory outlier detection on the spatial information. In this paper, a framework for congestion outlier detection with clustering method was proposed. Trajectory data are analyzed according to both temporal and spatial factors by introducing the concept of minimal bounding boxes (MBBs), and super dense clusters are regarded as congestion outliers. Experiments show the capability and efficiency of the proposed approach.