This study evaluates the in vitro accuracy and clinical repeatability of volumes derived by a new algorithm for three-dimensional reconstruction of cavity surfaces based on echocardiographic apical images obtained by probe rotation. The accuracy of the method was tested in latex phantoms (true volumes, 32 to 349 cm3) with (n = 9) or without (n = 9) rotational symmetry around the midcavitary long axis. Repeatability of left ventricular volumes was assessed in subjects without (n = 5) or with (n = 10) myocardial disease. Estimated phantom volumes obtained from four (three) imaging planes were close to true volumes with a mean difference +/- SD of 0 +/- 2 (2 +/- 3) cm3 in symmetric and 1 +/- 3 (4 +/- 4) cm3 in asymmetric objects. Biplane and single-plane volume estimates were less accurate. Interobserver and intraobserver repeatability of three-dimensional left ventricular volumes was good for analysis (coefficients of variation: 3.5% to 6.2%) and was lower for recording (coefficients of variation: 7.4% to 10.9%). Hence the present algorithm reproduces volumes of symmetric and deformed in vitro objects accurately over a wide range of size and shape, and it produces repeatable left ventricular volumes in the clinical situation.