Abstract: The free-form surface registration problem is important in medical image processing and reconstruction. An accurate, robust and fast solution is, therefore, of great significance. Most existing approaches, like Iterative Closest Point (ICP) or mutual information, can only work under many assumptions or limitations. In this paper we propose an new approach using an improved Genetic Algorithm [6] for finding the transformation, a translation and a rotation, between two free-form surfaces, that are partially overlapped. Application of the algorithm to 3D object model construction using multiple image integration is investigated. The experimental results showed that the improved genetic algorithm for surface registration and model construction is robust and is faster than all existing methods in the literatures.