Abstract: LIDAR data filtering is to obtain high-precision digital elevation model of improved hierarchical moving curved fitting algorithm, removed the rough points and most of the buildings side points, by gridding points, data index were calculated, used the least squares fitting surface, the adjacent surfaces maintained a certain overlap degree, automatically set the threshold for processing the raw data. Finally, used the multi-scale weighted interpolation method to generate digital elevation model. Adopt test data provided by international photogrammetry and remote sensing society(ISPRS).Results show that the algorithm reduced three types error possibility, total error of filtering in first study area decreases from 8.39% to 7.40%, and the total error in the second study area decrease from 6.39% to 5.86%, the generated digital elevation model to meet the precision requirements, algorithm can be popularized.