The effective detection and real-time tracking of vehicle in traffic surveillance video is the premise of vehicle behavior analysis and identification, and also the core content and key technology of intelligent transportation system (ITS). By analyzing the current key technologies of the vehicle attribute recognition method and vehicle video retrieval, combining with the own characteristics of traffic surveillance video itself and for the purpose of application, this paper designs a vehicle retrieval system which fuses the license plate, body color, vehicle models and other vehicle appearance attributes for hierarchical recognition. The system can provide users with a variety of search methods, so as to realize the accurate retrieval of relevant vehicles in traffic surveillance video.