A three-dimensional (3D) diffusion-weighted imaging (DWI) method for measuring cardiac fiber structure at high spatial resolution is presented. The method was applied to the ex vivo reconstruction of the fiber architecture of seven canine hearts. A novel hypothesis-testing method was developed and used to show that distinct populations of secondary and tertiary eigenvalues may be distinguished at reasonable confidence levels (P < or = 0.01) within the canine ventricle. Fiber inclination and sheet angles are reported as a function of transmural depth through the anterior, lateral, and posterior left ventricle (LV) free wall. Within anisotropic regions, two consistent and dominant orientations were identified, supporting published results from histological studies and providing strong evidence that the tertiary eigenvector of the diffusion tensor (DT) defines the sheet normal. Copyright 2005 Wiley-Liss, Inc.