A novel shape similarity retrieval algorithm (Hough-Radii) for 2-D objects is presented. The method uses a polar transformation of the contour points to get the shape descriptor that is invariant to translation, rotation and scaling. We take the maximum point in the generalized Hough transform (GHT) mapping array as the reference point for polar transform that is different from the traditional Centroid-Radii method where the geometric centre was taken as the origin. The effectiveness of our algorithm is illustrated in the retrieval of two databases of 99 and 216 shapes provided by Sebastian et al. The experimental results show the competitiveness of our approach to some others especially in the retrieval of partially occluded and missing images.