Reproducibility may be as important as absolute accuracy in assessing the utility of an echocardiographic method of left ventricular volume estimation for epidemiologic or physiologic studies. The magnitude of differences between measurements in the same subjects from day to day must be defined before any quantitative technique can be used reliably to document "real" changes in heart volume over time. Two-dimensional echocardiograms were performed serveral days apart in 30 subjects, including 20 normal subjects and 10 patients with stable coronary heart disease. Analyses of light-pen tracings provided measurements of end-diastolic volume, endsystolic volume and derived ejection fraction on both days, and differences in individual subjects between days were quantitated. Beat to beat, interobserver and intraobserver variability also were assessed. Although group values changed little from day to day, individual volume changes were substantial in some cases. Confidence limits for individual measurements were derived from analyses of intrasubject variability and were as follows: end-diastolic volume +/- 15%, end-systolic volume +/- 25%, ejection fraction +/- 10%. Confidence limits in a larger group of subjects were narrower; in a group of 30 subjects, changes of greater than 2% in end-diastolic volume, 5% in end-systolic volume and 2% in ejection fraction most likely represent real change. Intraobserver variability was minimal, but interobserver and beat to beat variability were of sufficient magnitude to suggest that serial measurements on a given subject be made ideally by a single person and that several cycles be averaged for a given measurement.