The thermodilution (TD) method of determining cardiac output (CO) is widely used in clinical practice, but its reliability in this setting is not well understood and may be poor. This is especially true when dealing with unstable patients and using commercial TD devices. Numerous published reports have shown markedly variable results when comparing simultaneously the TD method and the Fick or dye-dilution methods in catheterization laboratories, intensive care units, or operating rooms. In order to assist clinicians in interpreting the reliability of the TD method in measuring CO, we analyzed all available published data (14 reports). The differing results in these reports were evaluated by standard statistical methods and by an extension of the influence function method of structural analysis developed to differentiate reproducibility and accuracy errors of each technique. Comparing the accuracy of the TD method with that of the Fick or the dye-dilution methods reveals that the 3 methods are of equal merit and can be used as independent references. Our reproducibility data show that when using commercial TD devices there must be a minimal difference of 12 to 15% (average, 13%) between determinations of cardiac output (3 measurements per determination) to suggest clinical significance. Minimal differences of 20 to 26% (average, 22%) are required between determinations when using 1 measurement per determination. There was no difference in the quality of the TD method when comparing rapid injection of iced or room temperature thermal indicator.