Voxel-based morphometry (VBM) has detected differences between brains of groups of patients with epilepsy and controls, but the sensitivity for detecting subtle pathological changes in single subjects has not been established. The aim of the study was to test the sensitivity of VBM using statistical parametric mapping (SPM5) to detect hippocampal sclerosis (HS) and cortical neuronal loss in individual patients. T1-weighted volumetric 1.5 T MR images from 13 patients with HS and laminar cortical neuronal loss were segmented, normalised and smoothed using SPM5. Both modulated and non-modulated analyses were performed. Comparisons of one control subject against the rest (n = 23) were first performed to ascertain the smoothing level with the lowest number of SPM changes in controls. Each patient was then compared against the whole control group. The lowest number of SPM changes in control subjects was found at a smoothing level of 10 mm full width half maximum for modulated and non-modulated data. In the patient group, no SPM abnormalities were found in the affected temporal lobe or hippocampus at this smoothing level. At lower smoothing levels there were numerous SPM findings in controls and patients. VBM did not detect any abnormalities associated with either laminar cortical neuronal loss or HS. This may be due to normalisation and smoothing of images and low statistical power in areas with larger inter-individual differences. This suggests that the methodology may currently not be suitable to detect particular occult abnormalities possibly associated with seizure onset zone in individual epilepsy patients with unremarkable standard structural MRI. Hum Brain Mapp, 2009. © 2009 Wiley-Liss, Inc.