Determination of left ventricular (LV) volumes and ejection fraction (EF) from two-dimensional echocardiographic (2DE) images is subjective, time-consuming, and relatively inaccurate because of foreshortened views and the use of geometric assumptions. Our aims were (1) to validate a new method for rapid, online measurement of LV volumes from real-time three-dimensional echocardiographic (RT3DE) data using cardiac magnetic resonance (CMR) as the reference and (2) to compare its accuracy and reproducibility with standard 2DE measurements. CMR, 2DE, and RT3DE datasets were obtained in 50 patients. End-systolic and end-diastolic volumes (ESV and EDV) were calculated from the 2DE images using biplane method of disks. ES and ED RT3DE datasets were analysed using prototype software designed to automatically detect the endocardial surface using a deformable shell model and calculate ESV and EDV from voxel counts. 2DE and RT3DE-derived volumes were compared with CMR (linear regression, Bland-Altman analysis). In most patients, analysis of RT3DE data required <2 min per patient. RT3DE measurements correlated highly with CMR (r: 0.96, 0.97, and 0.93 for EDV, ESV, and EF, respectively) with small biases (-14 mL, -6.5 mL, -1%) and narrow limits of agreement (SD: 17 mL, 16 mL, 6.4%). 2DE measurements correlated less well with CMR (r: 0.89, 0.92, 0.86) with greater biases (-23 mL, -15 mL, 1%) and wider limits of agreement (SD: 29 mL, 24 mL, 9.5%). RT3DE resulted in lower intra-observer (EDV: 7.9 vs. 23%; ESV: 7.6 vs. 26%) and inter-observer variability (EDV: 11 vs. 26%; ESV: 13 vs. 31%). Semi-automated detection of the LV endocardial surface from RT3DE data is suitable for clinical use because it allows rapid, accurate, and reproducible measurements of LV volumes, superior to conventional 2DE methods.