Building 3D models of real-world objects by assembling views taken by a range sensor promises to be a more efficient method than manually producing CAD drawings. In this technique, a series of range images are acquired and then registered or aligned with each other to a high degree of accuracy. Finally, the polygonal meshes corresponding to the range images are merged to form a complete 3D model consisting of a single mesh. Many techniques have been proposed to solve the registration problem; however, little work has been done to date to compare several registration algorithms with the same sets of data. In this paper, we examine a software test-bed built for performing such comparisons. Within this test-bed, we have implemented several common registration algorithm variants to the baseline Iterative Closest Point (ICP) algorithm and tested them on partially overlapping range images taken from four different objects.