A new method is described for precise quantitative analysis of the relative three-dimensional distribution of myocardial tracers. The system uses a 360 degrees elliptical sampling of radial slices to create activity profiles. These are then positioned onto a common centre at the same angular coordinates as the corresponding radial slice reconstruction planes to generate a two-dimensional polar summary display. Abnormal distribution is then identified by automatic comparison of the patient polar map with the threshold of a normal database defined on a pixel by pixel basis as the normal mean -2.5 SD. Our stress and rest databases currently comprise 34 and 24 studies for sestamibi and tetrofosmin respectively. The present method differs from currently available software in two major respects. First, radial slices are used rather than short-axis slices to minimize operator intervention and to allow quantitative evaluation of the left ventricle volume independent of the heart size and without truncation, in particular near the apex and at the base. This sampling scheme also results in a more homogeneous and sampling-independent partial volume effect. Secondly, quantitative analysis is improved by calculating perfusion defect severity, extent and size in a precise manner. Severity is evaluated relative to a standardized background measurement and to the mean normal value rather than to the threshold value. This parameter was underestimated up to a defect extent of 32 cm2 in our phantom studies. Calculation of defect extent takes into account the surface distortion resulting from planar projection by using pixel by pixel weighted factors but it is otherwise overestimated as a result of the limited resolution of the imaging system. Integrating defect severity and extent, our hypoperfusion index appeared to accurately estimate the true defect size in our phantom model (r=0.993). The reproducibility of analysis was 6.24% in phantom studies and 3.10% in patient studies including repeated acquisitions. Applied to a well-documented population of 80 patients, this method resulted in an 86% sensitivity and a 78% specificity for overall coronary artery disease detection with reference to the angiographic data.