We present a crude registration algorithm for two partially overlapping range images acquired from a rigid object. Our proposal is characterized by two novel features: drastic data reduction and expressive shape descriptor. The data reduction is achieved on the basis of the observation that for crude alignments only the overall shape of image is of interest and the enhancement of the expression power of the shape descriptor is accomplished by integrating in it both local and global shape information. Experimental results show that our algorithm delivers good initial estimate to a fine registration program.