Deformation tensor morphometry makes use of the derivatives of spatial transformations between anatomies, to provide highly localized volumetric maps of relative anatomical size. The analysis of such maps, however, has the challenge of describing the data in a way that allows the spatial scale and extent of the local shape properties to match those induced by the disease process being studied. This study examines an approach to the spatial filtering of transformation Jacobian maps created in multisubject studies of brain anatomy, which constrains the filter neighborhood within common structural boundaries present in the spatially normalized image data. The filtering incorporates information derived from the spatial normalization process, using a statistical framework to introduce a measure of uncertainty in local regional intensity correspondence following spatial normalisation. The proposed filtering approach is compared to the use of spatially invariant Gaussian filtering in the analysis of Jacobian determinant maps of brain shape and shape change in Alzheimer's disease and normal aging. Results show significantly improved delineation of fine scale patterns of shape difference (in cross-sectional studies) and shape change (from multiple serial magnetic resonance imaging studies).