Abstract: A self-adaptive filtering algorithm for scattered points was proposed. The node splitting algorithm and the clustering algorithm of R*-tree were improved and the spacial index structure of triangular mesh model was established based on the improved R*-tree; The local surface reference data was obtained according to data nodes- distributing of the spacial index structure; The filtering weight was computed by self-adjusting figure feature of 2-D Gauss distributing according to its local surface reference data; The weight value of local surface reference data to originality surface data was computed, and the coordinate of scattered points was computed according to this weight value; The self-adaptive filtering for scattered points was realized with the method of weighted mean. It proved that this algorithm can improve the efficiency of filtering and reduce the random error of the scattered points on the basis of the accurate reservation of surface characteristic.