In this paper we present a method for the adapting a geometrical deformable model (GDM) for reconstruction of real-world objects from multiple range images. Our approach registers the range images simultaneously, carves out an immediate volume and finally generates an accurate, sparse triangle mesh. The proposed GDM scheme refines an initial roughly approximated mesh by deformation and adaptive subtriangulation. Even in the case of very large data sets our approach presents an efficient method of surface reconstruction due to adaptive improvement to the desired degree of accuracy. Since the root mean square approximation error of each triangle is minimized in an iterative procedure, the mesh quality is higher than that of previous approaches.