A method for building model data for CAD and CAM purposes from physical instances using three-dimensional sensor data is presented. These techniques are suitable for Reverse Engineering of industrial parts, and can be used as a design aid as well. The nature of the reverse engineering task is quantitative, and the emphasis is on accurate recovery of the geometry of the part, whereas the object recognition task is qualitative, and aims to recognize similar shapes. The proposed method employs multiple representation to build a CAD model for the part, and to produce useful information for part analysis and process planning. The model building strategy is selected based on the obtained surface and volumetric data descriptions and their quality. A novel, robust non-linear filtering method is presented to attenuate noise from sensor data. Volumetric description is obtained by recovering a superquadric model for the whole data set. A surface characterization process is used to determine the complexity of the underlying surface. A substantial data compression can be obtained by approximating huge amount sensor data by B-spline surfaces. As a result a Boundary Representation model for Alpha-1 solid modeling system is constructed. The model data is represented both in Alpha-1 modeling language and IGES product data exchange format. Experimental results for standard geometric shapes and for sculptured free-form surfaces are presented using both real and synthetic range data.