The aim of this study was to compare three automatic programs (P1, P2, P3) for evaluating radionuclide left ventricular ejection fraction (LVEF) and to emphasize the clinical consequences. Gated radionuclide ventriculography was performed in 73 subjects, 15 of whom were healthy and 58 of whom had experienced heart failure. All scintigraphic data were processed with the three programs. Good inter-observer, intra-observer and automatic-manual reproducibility were observed using each of the three programs. On the other hand, in the normal subjects, the three mean normal LVEF values were significantly different from each other (P1 = 77 +/- 5%, P2 = 63 +/- 7%, P3 = 68 +/- 8%; P < 0.0001) In the pathological patients, the values obtained with P2 were significantly different from those obtained using P1 and P3 (P1= 32 +/- 15%, P2 = 26 +/- 13%; P < 0.0001), Moreover, the linear regression studies between the three automatic programs were always significantly different from the identity line equation (y = x). This study shows that LVEF criteria for normality depend on the program used, and inter-program measurement of LVEF is poorly reproducible. Caution is recommended when comparing data obtained from different centres (or different computers), either in the follow-up of a given patient or in gathering results from patient groups.