In patients with drug-resistant focal epilepsies subdural grid electrodes may be implanted to determine the seizure onset zone and eloquent cortex areas. Since the spatial relationship of the grid to the underlying brain is poorly visualized on MRI, we co-registered MRI before and CT after implantation of subdural grid electrodes. In this study we sought an appropriate algorithm to combine both imaging modalities. We compared six different co-registration algorithms including surface-oriented, mutual information-based and landmark-based methods. The resulting overlay matrices were analyzed by calculating rotational and translational shifts and by judging co-registered MRI and CT scans visually. A brain surface oriented method had the lowest rotational (axial 0.7 +/- 0.6 degrees; coronal 1.7 +/- 1.1 degrees; sagittal 1.9 +/- 1.8 degrees) and translational shifts (3.7 +/- 1.3 mm). It was judged visually to be the best, had a low intra- and inter-observer variability, and lasted approximately 15 minutes. This algorithm is recommended when co-registering MRI before and CT after implantation of subdural grid electrodes. Skin-, voxel-, and landmark-based algorithms are less accurate, which is most likely due to postsurgical deformation of extra- and intracranial soft tissue.