Abstract: This paper addresses the peg-in-hole problem, and presents an active vision strategy which includes three relative algorithms: (1)3D location estimation of circular feature by active vision;(2)3D location estimation of straight line by active vision;(3)camera pose estimation using one feature point. By setting up an object-centered coordinate, the relative position between robot hand and the object can be measured precisely even in the presence of intensive uncertainties of cameras and robot. Experimental result on a real robot system shows that this method is valid. This method can be extended to other forms of robot operations.