Left ventricular mass (LVM) measurements made by the truncated ellipsoid algorithm from clinical two-dimensional echocardiograms (2DE) were compared to autopsy weights in 37 patients. All six 2DE instruments were calibrated with an ultrasound phantom to standardize LVM measurements. Measurements were made by an experienced echocardiographer (LVME) and by an echocardiographer (LVMN) newly trained in LVM measurement from clinical 2DE tapes of patients with LV weights later confirmed at autopsy. LVME (r = 0.91, SEE +/- 41 gm) were more accurate than LVMN for all 2DE, but LVMN equalled LVME in accuracy for technically good 2DE. Interobserver variability was 36 gm, or 17% of LVM for all 2DE, and fell to 27 gm, or 12% of LVM for technically good 2DE. Segmental wall motion abnormalities and time from 2DE to death did not influence measurement accuracy significantly. LVM measurements by the 2DE truncated ellipsoid formula are accurate and reproducible in patients with normal and abnormal hearts.