Studies of memory-impaired patients will be most useful when quantitative neuroanatomical information is available about the patients being studied. Toward that end, in the case of medial temporal lobe amnesia, protocols have been developed from histological material that identify the boundaries of relevant structures on magnetic resonance images. Because the size of these structures varies considerably in the normal population, some correction for overall brain size is usually employed when calculating volume measurements. Although different correction procedures have been used to normalize for brain size, there has been little study of how well different methods reduce variability and which methods might be most useful. We measured the volume of the hippocampal region (hippocampus proper, dentate gyrus, and subicular complex) and the volumes of the temporopolar, entorhinal, perirhinal, and parahippocampal cortices in five memory-impaired patients and 30 controls. We then compared three different methods for normalizing the volume measurements: normalization by intracranial volume, normalization by aligning the brain to a standard atlas, and normalization by brain area at the level of the anterior commissure. Normalization by intracranial volume reduced variability in the volume measurements of nearly all brain regions to a greater extent than did normalization by other methods. When normalized by intracranial volume, the patients exhibited a mean reduction in hippocampal volume of about 40% and negligible reductions in the volumes of other medial temporal lobe structures. On the basis of earlier histological analysis of two other patients (L.M. and W.H.), who also had reductions in hippocampal size of about 40%, we suggest that a volume reduction in this range likely indicates a nearly complete loss of hippocampal neurons.