Abstract: To handle the challenging problem for inferring the upright orientation of 3D models only from geometry,we propose an automatic upright orientation detection algorithm for man-made objects in this paper.The proposed method takes advantages of the orientation clues from the shape properties,such as the facet orientation,symmetry and 3D convex hull facet orientation.We first use these clues to extract candidate orientations and get those orientation triplets that the orientations are pairwise orthogonal to form a frame.Next,we rotate the frame to the canonical coordinate and select the best one that aligns the model with canonical coordinate using model facet normal and area.Finally,we use various criterions on static stability and visibility to choose the correct upright orientation from the six axis-aligned candidates of canonical coordinate orientation.We have tested the proposed method on a 3D model database,and the results show that the proposed method outperforms the state-of-the-art methods.