Computational models of cardiac tissue electrophysiology have proved invaluable in studying control of cardiac function (1). They require high resolution models of tissue geometry with anisotropic and orthotropic architecture (2), with voxel dimensions of less than 0.2 mm for ventricular simulations and 0.05-0.1 mm for atrial and nodal simulations. DT-MRI (3) can provide such datasets from post mortem samples. A fixed (10% formal saline) and immobilised (polyacrylamide gel) rabbit entire heart was imaged using a Bruker 9.4 T MR instrument and a standard reduced encoding diffusion-weighted gradient echo sequence at 20°C: 500 ms TR; 18.7 ms TE; diffusion gradients with 5 ms duration and 8.9 ms separation; b = 1130 s/mm². Voxel resolution was 0.2 mm³. For each scan we acquired a single b0 image and diffusion-weighted images sensitised in each of 12 optimised directions (4). Single scan time for one heart was ~10 hours. We performed 6 scans per heart and calculated mean measurements at each voxel. Each heart gives a dataset of ~300 MB, and is available from us on request. The three eigenvalues and corresponding eigenvectors of the diffusion tensors were calculated at each voxel throughout the dataset. The primary eigenvector provides a measure of fibre orientation throughout the atria and ventricles, and the secondary eigenvector an index of any ventricular sheet structure – see Fig. 1. The resolution of the datasets enables the rotational fibre structure and the sheet structure to be mapped across the ventricular walls, and fibre orientations can be distinguished in the thinner atrial walls. DT-MRI provides an efficient method to build libraries of high-resolution structural models on which to run electrophysiology simulations (see Aslanidi et al., this Proceedings volume).