In this paper we present a contour-based multisensor image registration scheme which uses region boundaries and other strong edges as matching primitives. Chain code correlation and other shape similarity criteria such as moments are used to match closed contours. For the open contours, salient segments such as corners are detected first and then used in the matching process. Experimental results on several sets of Landsat, SPOT and Seasat images indicate that this approach is quite robust in matching the images with large rotation.