The difficulties associated with the experimental determination of high-resolution, three-dimensional (3D) structures of G-protein-coupled receptors (GPCRs), make molecular modeling the only practicable technique to provide structural information at atomic resolution of these integral membrane proteins. However, these methods need to be critically evaluated to assess their predictive capabilities and their limitations. We have recently proposed a de novo method to construct 3D structures of the transmembrane domain of GPCRs. The procedure requires only the knowledge of the amino acid sequence and uses the helix spatial arrangement of rhodopsin deduced from its low-resolution electron density map. Models are refined using molecular dynamics. The aim of the present work is to apply this approach to construct and refine a 3D model of rhodopsin as a test case. Moreover, to assess the level of accuracy that can be expected of the models using our de novo approach, we constructed and refined a 3D model of bacteriorhodopsin using a low-resolution electron density map of this protein. After refinement the model of bacteriorhodopsin was compared with the experimental structure refined at 3.5 Angstrom resolution.