A robust method to derive hypotheses for planar surfaces of 3D objects from noisy contours, is reported. It is a step to establish 3D scene model and a part of general vision system. A key concept is the fact that the object contour, which forms the geometrical shape of a 3D object, should be connected and that the connection forms a hypothesis for a plane which can be a part of a planar surface. The method works as follows; 1)Make a“local”plane from two line segments that have high connectivity. 2)Group local planes that have good coplanarity into a plane hypothesis. 3)Add the remaining coplanar segments into plane hypotheses. Results from synthetic data are presented and applications to the establishment of planar surfaces of objects are discussed.