According to the characters of web news stream, an on-line news event detection (ONED) method, based on the two-stage clustering, is proposed to solve the problem of repeated matching. A novel incremental event model was established by calculating terms weighting of events directly. Two stages are involved in our method. In the first stage, the similar reports collected in a certain period were clustered into micro-clusters. In the second, the micro-clusters were matched with existed events, and then this method updated the event model. Experiment shows that the proposed method improves the efficiency and accuracy of ONED with lower complexity and less feature information loss. 