Several multivariate statistical models have recently been introduced for estimation of left ventricular mass from standard 12-lead electrocardiographic measurements. The validity of these algorithms has not been adequately evaluated. The objective of this investigation was to compare the associations between echocardiographic and electrocardiographic left ventricular mass values with clinical and subclinical indexes of coronary heart disease. The evaluation was performed with participants of the Cardiovascular Health Study, a population-based sample of 5201 men and women aged 65 years and older. Echocardiographic M-mode measurements of left ventricular mass were performed from videotape recordings with the use of a strictly standardized protocol. Electrocardiographic algorithms of the Novacode program and new algorithms derived from the Cardiovascular Health Study population were used for left ventricular mass prediction. Echocardiographic and electrocardiographic determinations of left ventricular mass were technically successful in 3410 (65.6%) and 5013 (96.4%) participants, respectively. The Novacode model overestimated echocardiographic left ventricular mass. Compared with the Novacode model, the new Cardiovascular Health Study electrocardiographic model, which includes adjustment for body weight, eliminated left ventricular mass prediction bias and improved the correlation between echocardiographic and electrocardiographic left ventricular mass from .33 to .54 in women and from .46 to .51 in men. Echocardiographic and electrocardiographic models both demonstrated similar and about equally strong associations with overt and subclinical disease and with risk factors for left ventricular hypertrophy. These observations demonstrate the potential utility of electrocardiographic models for left ventricular mass estimation.