Reorientation of transaxial cardiac PET images into short-axis images has been shown by other investigators to improve visual identifiability of defects in myocardial tracer uptake. However, quantification of physiologic processes from such reoriented images may be complicated by errors introduced during the reorientation process. Therefore, a quantitative characterization of these errors is necessary. An annular phantom of human cardiac dimensions was imaged in a 15-plane positron emission tomograph at six angles (0 degrees, 5 degrees, 25 degrees, 45 degrees, 65 degrees, 85 degrees) and at two different axial sampling densities. In addition, two different reorientation interpolators were employed, one using three-dimensional linear interpolation and the other using a "hybrid" interpolation algorithm. Distortion of linear distances was variable but was minimized with denser axial sampling and the use of hybrid interpolation. Circumferential profile analysis, corrected for inhomogeneities in reoriented image spatial resolution, revealed a maximal loss of region of interest counts at 65 degrees of at least 14.4%. Reorientation errors were minimized by use of dense axial sampling, low angles of reorientation and use of the hybrid interpolation algorithm.