The purpose of this study was to investigate the relationship between three-dimensional (3D) deformation patterns in the canine left ventricle and localized graded reductions in perfusion. Magnetic resonance (MR) tissue tagging in a clinical scanner was used to determine systolic 3D deformation throughout the left ventricle with 32-msec time resolution. Six dogs were studied at normal and reduced left anterior descending coronary artery flow levels, for a total of 14 studies. Deformation was calculated by fitting a 3D displacement field to tag displacement data from three orthogonal sets of tags and taking spatial derivatives. A novel index of 3D radial mechanical function, calculated from the 3D strain tensor components and the tissue incompressibility constraint, had a higher correlation (R = 0.94) with perfusion (colored microspheres) than any of the 3D Lagrangian finite strain tensor components or wall thickening. As a function of the fraction of baseline perfusion, it was well fit by a linear relationship for subnormal perfusion with a slope of 0.46 +/- 0.05 and an intercept of -0.156 +/- 0.026. Longitudinal strain was lost first with decreasing perfusion (48%), followed by circumferential (40%) and finally radial function (35%). The strain method detected perfusion drops as small as 20%, and early paradoxical strain transients lasting 100 msec were seen only with ischemia. 3D strain changes can be noninvasively measured throughout the left ventricle with MR tissue tagging. MR imaging-derived strain indices, unique to 3D analysis, correlate most sensitively with regional perfusion in the canine left ventricle.