An important aspect of reverse engineering is the production of digital representations of physical objects for CAD systems. The first stage involves taking 3D coordinate measurements for points on the surface of the object and producing in general an unstructured set of points, called a point cloud. A triangulated surface can be generated from such a point cloud, allowing copies of the original object to be manufactured. However, these triangulated surfaces generally consist of a very large number of triangles with small errors in the positions of their vertices. In many cases the original object is made up of parts of a number of simple geometric objects. Our aim is to segment the triangulated surface into a small number of components, each of which approximates to part of a simple geometric shape. We have developed algorithms for curvature estimation in order to support a 'region growing' method of segmentation.