In this paper, we first show that the model selection is a vital part of the segmentation of multi-structural data. We then proceed by redefining the range segmentation problem as an instance of geometric computation and devise a new robust algorithm based on random sampling and rank ordering statistics. A novel characteristic of the presented algorithm is that it performs the model selection and segmentation simultaneously. This feature is particularly important for segmenting multi-structural data where the underlying functions of distinct segments have different orders. Furthermore, the method we employ for estimating the scale of the noise and distinguishing inliers from outliers is essentially different from those methods using least K-th order statistics. Finally, we present the result of applying our algorithm to a range segmentation problem.