There is wide beat-to-beat variability in cycle length and left ventricular performance in patients with atrial fibrillation. In this study, left ventricular ejection fraction and relative left ventricular volumes were evaluated on a beat-to-beat basis with the computerized nuclear probe, an instrument with sufficiently high sensitivity to allow continuous evaluation of the radionuclide time-activity curve. Of 18 patients with atrial fibrillation, 5 had mitral stenosis, 6 had mitral regurgitation, and 7 had coronary artery disease. Fifty consecutive beats were analyzed in each patient. The mean left ventricular ejection fraction ranged from 17 to 51%. There was substantial beat-to-beat variation in cycle length and left ventricular ejection fraction in all patients, including those with marked left ventricular dysfunction. In 14 patients who also underwent multiple gated cardiac blood pool imaging, there was an excellent correlation between mean ejection fraction derived from the nuclear probe and gated ejection fraction obtained by gamma camera imaging (r = 0.90). Based on beat-to-beat analysis, left ventricular function was dependent on relative end-diastolic volume and multiple preceding cycle lengths, but not preceding end-systolic volumes. This study demonstrates that a single value for left ventricular ejection fraction does not adequately characterize left ventricular function in patients with atrial fibrillation. Furthermore, both the mean beat-to-beat and the gated ejection fraction may underestimate left ventricular performance at rest in such patients.