The aim of this preliminary study was to evaluate the accuracy of left and right ventricular output computed from a semi-automatic processing of tomographic radionuclide ventriculography data (TRVG) in comparison with the conventional thermodilution method. Twenty patients with various heart diseases were prospectively included in the study. Thermodilution and TRVG acquisitions were carried out on the same day for all patients. Analysis of gated blood pool slices was performed using a watershed-based segmentation algorithm. Right and left ventricular output measured by TRVG correlated well with the measurements obtained with thermodilution (r = 0.94 and 0.91 with SEE = 0.38 and 0.46 l/min, respectively, P < 0.001). The limits of agreement for TRVG and thermodilution measurements were -0.78-1.20 l/min for the left ventricle and -0.34-1.16 l/min for the right ventricle. No significant difference was found between the results of TRVG and thermodilution with respect to left ventricular output (P = 0.09). A small but significant difference was found between right ventricular output measured by TRVG and both left ventricular output measured by TRVG (mean difference = 0.17 l/min, P = 0.04) and thermodilution-derived cardiac output (mean difference = 0.41 l/min, P = 0.0001). It is concluded that the watershed-based semi-automatic segmentation of TRVG slices provides non-invasive measurements of right and left ventricular output and stroke volumes at equilibrium, in routine clinical settings. Further studies are necessary to check whether the accuracy of these measurements is good enough to permit correct assessment of intracardiac shunts.