The intrinsic variability and accuracy of left ventricular ejection fraction determined by multiple gated cardiac blood pool imaging was evaluated in 83 patients. Ejection fraction by gated studies correlated well with data from first pass radionuclide angiocardiography (r = 0.94) and from contrast angiography (r = 0.84). Intra- and interobserver variabilities of absolute ejection fraction were minimal (mean +/- standard deviation 1.4 +/- 1.2 and 1.6 +/- 1.5 percent, respectively) and were not different for normal (ejection fraction 55 percent or greater) and abnormal patients. Ejection fraction was determined twice in 70 patients: on the same day at intervals separated by 1 to 2 hours (41 patients) and on 2 different days (29 patients). Ejection fraction ranged from 18 to 91 percent and was normal in 37 patients. There was no difference in mean serial variabilities of absolute ejection fraction for all repeat studies performed on the same and separate days (3.3 +/- 3.1 versus 4.3 +/- 3.1 percent (not significantly different). The mean variability of absolute ejection fraction for repeat studies in normal patients was significantly greater than in abnormal patients (5.4 +/- 4.4 versus 2.1 +/- 2.0 percent, P less than 0.01). The incidence rate of absolute interstudy changes of 5 percent or more was significantly higher in normal than in abnormal patients (P less than 0.01). This differential variability should be considered in interpreting sequential changes in left ventricular ejection fraction. To be attributed to nonrandom physiologic alterations, the absolute change in ejection fraction should be 10 percent or more in normal patients and 5 percent or more in abnormal patients.