Abstract: The vehicle trajectory data contains macroscopic information about the behavior of urban traffic and moving objects, from which valuable city traffic trends and vehicle behavior patterns and other informa- tion can be discovered. Analyzing trajectory data is important for traffic management. In response to the ir- regular vehicle trajectory data and the lack of an accurate description of group behaviors, this paper proposes a density-based trajectory clustering method. Our method partitions the trajectory data into segments ac- cording to angle and distance. The similarity of segments is measured with a new trajectory distance function. Representative trajectories are generated for the clustering results. Our experimental results based on three trajectory datasets demonstrate that the representative trajectories generated from our proposed method pro- vide a better description for the overall trend of traffics, which could better serve the traffic management.