Abstract: A robust hybrid segmentation method for quadric feature extraction is proposed. It is composed of three steps. First, points are segmented under the controlled error of deviation from a certain surface type. Second, the points with similar geometric feature are extracted from the points segmented in the first step. Finally, quadric feature extraction is implemented using a connectivity criterion for point sets. Experimental results show that the precision of reconstructed model and efficiency of remodeling are improved remarkably.