Abstract: It relies greatly on the accident prediction models to make effective traffic safety countermeasures. Therefore, by taking Harbin urban arterial network composed of 468 arterial links and 163 at-grade intersections as a case, the broad geometry and traffic flow data of the network were collected, as well as 8 510 accident data occurred on the network during 1999 to 2004. Firstly, the characterist ics of the traffic accident data were analyzed, and the results show that theaccident data follow the Negative Binomial distribution. Secondly, links and inter sections were classified according to the cluster analysis method, and then the accident prediction models that can be used to predict the accident frequencies occurred on each kind of links and intersections were established. Thirdly, the quantitative relationship between the accident index of the links during rushh ours and the v/c of them was discussed. Totally, 24 prediction models were calibrated. Finally, the prediction models were applied to a case study on partial road network of Harbin, which was planned for the target year of 2010. There sults show the fact that the accident prediction models are effective.