For accurate electroencephalogram-based localization of mesial temporal and frontal sources correct modeling of skull shape and thickness is required. In a simulation study in which results for matched sets of computed tomography and magnetic resonance (MR) images are compared, it is found that errors arising from skull models based on smooth and inflated segmented MR images of the cortex are of the order of 1 cm. These errors are comparable to those found when overestimating or underestimating skull conductivity by a factor of two.