Due to the limited spatial resolution of positron emission tomography (PET), the accuracy of quantitative measurements of regional metabolism or neuroreceptor concentration is influenced by partial volume averaging of brain with CSF, bone, and scalp. This effect is increased in the presence of cortical atrophy, as in patients with Alzheimer disease (AD). Correction for this underestimation in PET measurements is necessary for the comparison of AD patients and normal controls. We have developed a method for three-dimensional correction of human PET data using magnetic resonance (MR) imaging. A composite brain tissue image is created by summing the binary representation of nine MR images, weighted to the PET z-axis line-spread function. This composite tissue image is convolved to the resolution of the PET image. The original PET image is divided by the convolved tissue image on a pixel-by-pixel basis, resulting in an atrophy-corrected PET image in which count density represents activity per volume of brain tissue rather than spatial volume. This has been performed in [11C]carfentanil mu-opiate receptor PET studies of the temporal cortex in two AD patients and one normal volunteer. After correction, average regional increases in count density were 11% (range = 4-21%) in the normal and 46% (range = 28-99%) and 48% (range = 14-109%) in the patient studies. The accuracy of this method of partial volume correction was estimated using a spherical phantom.