This study describes a method of automatic border detection of the left ventricular endocardium and epicardium associating three methods of segmentation (increase of region, border detection and adaptive threshold), applicable to the evaluation of ventricular mass and volume by magnetic resonance imaging. Despite slight underestimation, the spin-echo sequence used in 9 small pigs provided a value of left ventricular mass close to that observed ex vivo (r = 0.97, SEE = 6.05 g). Clinical validation using a rapid gradient-echo sequence was undertaken and compared with manual border detection carried out by three independent, trained operators. The study population included healthy subjects and patients with global or segmental left ventricular dysfunction with or without ventricular deformation. The correlations between automatic and manual detection were satisfactory both for calculation of left ventricular mass (r = 0.93, SEE = 13 g) and measurement of surfaces (r = 0.91, SEE = 1.4 cm2). The concordance of the two methods was confirmed by the Bland and Altman test. Cardiac magnetic resonance imaging may provide accurate and reproducible measurements of left ventricular mass within acceptable acquisition and image processing times for routine use. Although the clinical value of such a method is accepted both for establishing the prognosis and guiding management, studies of the cost/efficacy ratio should be undertaken to situate magnetic resonance imaging with respect to other non-invasive techniques of investigation of left ventricular function.