The accuracy of measuring cardiac stroke volume with a new transesophageal phased array ultrasonic probe was investigated in 10 dogs. The method involved scanning the heart to obtain serial images covering the entire left ventricular cavity at end-expiration. An off-line computer analysis of the images was used to form three-dimensional reconstructions of the left ventricular cavity at end-diastole and end-systole, from which stroke volume was determined. Comparison with stroke volume determined by thermodilution during a wide range of hemodynamic conditions gave the following results for least-squares regression on 57 determinations (with the 95% confidence limits in parentheses): slope 0.95 (0.842-1.06), ordinate intercept 2.1 (2.0-2.2) ml, standard error of the estimate 4.1 ml, and correlation coefficient of 0.92 (0.87-0.95). Histologic examination of sections of esophagus surrounding the tip of the probe in nine dogs demonstrated minimal trauma to the esophageal wall, with eight specimens described as normal and one showing mild inflammation. The authors conclude that three-dimensional reconstruction of the left ventricular cavity from multiple transesophageal images offers a safe and accurate, although presently tedious, method for determining stroke volume.