In order to investigate the possibility of regional variation of ventricular structure, 25 normal postmortem human hearts were studied by inspection of cavity shape and subepicardial fibre orientation, by dissection, and by the histology of sections in two orthogonal planes. Ventricular architecture was complex. Inlet and outlet long axes were separated by 30 degrees in the left ventricle. In the right the corresponding figure was 90 degrees. The thickest part of the left ventricular wall was at the base. At the apex there was potential endo- and epicardial continuity. Left ventricular cavity shape departed significantly from any simple geometric figure, there being, consistently, regions of both positive and negative curvature on the diaphragmatic aspect. The presence of trabeculae caused considerable variation in wall thickness. Striking variation was found in the arrangement of subepicardial muscle fibres. Most pronounced was the contrast between the longitudinal arrangement of fibres observed on the oblique margin and the circumferential arrangement of those on the acute. On the diaphragmatic surface of the left ventricle, fibres near the crux and apex ran circumferentially while those between ran obliquely; those on the diaphragmatic surface of the right ventricle also ran circumferentially. Deeper in the myocardium the arrangement was simpler. In the mid-wall of the left ventricle fibres were circumferential, best developed towards the base and in the upper part of the septum. Near the apex of the left ventricle and in the mid-wall of the right ventricle such fibres were sparse. The subendocardial region consisted of longitudinally directed fibres forming the trabeculae and papillary muscles, while fibres deep to and between the trabeculae coursed more obliquely. These findings were confirmed by histology. Models based on uniform myocardial fibre structure cannot explain wall movement in normal subjects, and are likely to have significant limitations if used to investigated left ventricular function in disease.