Fire statistics and fire analysis have become important ways for us to understand the law of fire,prevent the occurrence of fire, and improve the ability to control fire. According to existing fire statistics, the weighted fire risk calculating method characterized by the number of fire occurrence, direct economic losses,and fire casualties was put forward. On the basis of this method, meanwhile having improved K-mean clus-tering arithmetic, this paper established fire dsk K-mean clustering model, which could better resolve the automatic classifying problems towards fire risk. Fire risk cluster should be classified by the absolute dis-tance of the target instead of the relative distance in the traditional cluster arithmetic. Finally, for applying the established model, this paper carded out fire risk clustering on fire statistics from January 2000 to December 2004 of Shenyang in China. This research would provide technical support for urban fire management.