The interaction between osteoarthritis (OA) pain and brain properties remains minimally understood, although anatomical and functional neuroimaging studies suggest that OA, similar to other chronic pain conditions, may impact as well as partly be determined by brain properties. Here, we studied brain gray matter (GM) properties in OA patients scheduled to undergo total joint replacement surgery. We tested the hypothesis that brain regional GM volume is distinct between hip OA (HOA) and knee OA (KOA) patients, relative to healthy controls and moreover, that these properties are related to OA pain. Voxel-based morphometry group contrasts showed lower anterior cingulate GM volume only in HOA. When we reoriented the brains (flipped) to examine the hemisphere contralateral to OA pain, precentral GM volume was lower in KOA and HOA, and 5 additional brain regions showed distortions between groups. These GM changes, however, did not reflect clinical parameters. Next, we subdivided the brain into larger regions, approximating Brodmann areas, and performed univariable and machine learning-based multivariable contrasts. The univariable analyses approximated voxel-based morphometry results. Our multivariable model distinguished between KOA and controls, was validated in a KOA hold-out sample, and generalized to HOA. The multivariable model in KOA, but not HOA, was related to neuropathic OA pain. These results were mapped into term space (using Neurosynth), providing a meta-analytic summary of brain anatomical distortions in OA. Our results indicate more subtle cortical anatomical differences in OA than previously reported and also emphasize the interaction between OA pain, namely its neuropathic component, and OA brain anatomy.