This paper addresses an approach of segmenting dense range. It is the most difficult and important step of constructing a solid model from dense range obtained from 3D scanner. The original object is assumed to be covered by piecewise quadric surfaces and planes, and the range image can be stratified in the way that the points on different layers are situated on parallel planes respectively. The method is based on the strategy "from one dimension manifold to two dimension manifold". Quadric curve segments and straight line segments on every layer is recognized firstly. Group the quadric curve and straight line segments situated on a sequence of adjacent parallel planes with some coherent properties. The quadric surfaces or planes bounding the original object are composed by the quadric curve and straight line segments in same groups. Three examples show the algorithm's performance. This approach can be applied in reverse engineering to construct solid model of mechanical parts.