We propose a new algorithm for automatically computing approximations of a given polyhedral object at different levels of details. The applications for this algorithm is the display of very complex scenes. Our approach is similar to the region-merging method used in image segmentation. We iteratively collapse edges based on a measure of the geometric deviation from the initial shape. when edges are merged in the right order, this strategy produces a continuum of valid approximations of the original object, which can be used for faster rendering at vastly different scales.