This thesis presents the development and test of an algorithm to evaluate the data provided by an optical profile measurement system. It is suitable for all profile cross-sections, including both open as well as closed profiles. The data, generated by two laser triangulation sensors, is automatically evaluated and compared with the information from a technical drawing. The profile contour depicted by the measurement data doesn't necessarily have to be continuous, due to the complex profile cross-sections and the associated possible occlusion of the sensors. This is a non-registration problem, there may be no affine mapping between the measured and the CAD-data, due to possible deformations of the produced profiles. The evaluation algorithm enables a complete and automatic geometric quality inspection of the manufactured profiles. Furthermore, a better understanding of the interaction of material properties respective process parameters and the required quality characteristics can be gained. The core of this software is a generally applicable segmentation algorithm, which is able to separate a contour consisting of measurement data points into the corresponding line and circle elements. The first step is a rough segmentation based on the principle of two-dimensional cross-correlation, followed by an iterative fine segmentation. The geometric properties of the individual elements are then computed with a contour fit, first without and then including boundary conditions, defined with respect to a-priori knowledge of the manufacturing process. During the development, special care was taken on minimizing the required computing power so that the algorithm is suitable for series evaluation. This is done on the basis of previous calculations of profile cross-section specific information with the aid of the CAD drawing and, on the other hand, through an efficient program structure. The algorithm was successfully applied to perform reverse engineering on manufactured parts of a rolling mill.