Two classes of functional neuroimaging methods exist: hemodynamic techniques such as PET and fMRI, and electromagnetic techniques such as EEG/ERP and MEG. In order to fusion these images with anatomical information, co-registration with volumetric MRI is needed. While such co-registration techniques are well established for hemodynamic images, additional steps are needed for electromagnetic recordings, because the activity is only recorded on the scalp surface and inverse solutions based on specific head models have to be used to estimate the 3-dimensional current distribution. To date most of the experimental and clinical studies use multi-shell concentric sphere models of the head, solve the inverse problem on this simplistic model, and then co-register the solution with the MRI using homogeneous transform operations. Contrary to this standard method, we here propose to map the MRI to the spherical system by defining transformation operations that transform the MRI to a best-fitting sphere. Once done so, the solution points are defined in the cerebral tissue of this deformed MRI and the lead field for the distributed linear inverse solutions is calculated for this solution space. The method, that we call SMAC (Spherical Model with Anatomical Constrains) is tested with simulations, as well as with the following real data: 1) estimation of the sources of visual evoked potentials to unilateral stimulation from data averaged over subjects, and 2) localization of interictal discharges of two epileptic patients, one with a temporal, the other with an occipital focus, both confirmed by seizure freedom after resection of the epileptogenic region.