SUMMARY The accuracy of two-dimensional echocardiographic left ventricular volume measurement in an isolated heart preparation was tested using Simpson's reconstruction of progressively fewer short-axis cross sections of known location. Echocardiographic images from five ejecting hearts submerged in a special tank were obtained under conditions designed for maximal accuracy of echocardiographic volume assessment. Echocardiographic determinations of 52 volumes at various times throughout the cardiac cycle were compared, by least-squares linear regression, with simultaneous direct-volume measurements by volumetric chamber (range 9.4-44.8 ml). Echocardiographic and direct measurements correlated well for all numbers of cross sections from 1-19 (r = 0.84-0.97); however, variability of direct volume predicted from a given echocardiographic measurement increased nonlinearly as the number of cross sections per heart decreased, and was especially large when three or fewer cross sections were used (SEE = 4.6-7.1 ml). We previously used isolated, ejecting dog hearts to compare directly measured ventricular volume (without reliance on geometric assumptions) with echocardiographic volume obtained by reconstructing multiple cross-sectional images obtained at 3-mm intervals along the vertical axis of the heart.'0 Comparison of simultaneous data obtained throughout the cardiac cycle revealed not only a high correlation of echo to direct volume, but also a high predictive value of direct volume from any echocardiographic determination owing to low variability (r = 0.972, SEE = 2.93 ml, direct volume range 9.4-54.7 ml). These measurements were obtained under highly controlled conditions and it was not expected that the accuracy of echocardiographic volume determination could be matched in clinical use or in research applications in intact animals. At the very least, one could not obtain the large number of slices used in that study, and we did not assess the accuracy of volume changes with decreasing numbers of slices.Using the same preparation in the present study, but analyzing the data using progressively fewer slices, we found that at least four cross-sectional images of known location are required to predict ventricular volume without significant loss in accuracy.