The objective of this study was to test the value of electrocardiogram for predicting left ventricular mass (LVM), assessed echographically in 136 asymptomatic men with at least one major cardiovascular risk factor. We measured the Sokolow-Lyon and Cornell voltages, as well as the ratio of Cornell voltage to QRS voltage in lead II. The prevalence of left ventricular hypertrophy (LVH), defined as LVM of =< 125 g/m 2 , was 6%, whereas that of increased LVM, defined as LVM of =< 99 g/m 2 , the 90th upper percentile of a control group, was 29%. Receiver operating characteristics curves showed that for predicting LVH at 80% specificity, the Cornell/QRS I I voltage ratio had a sensitivity of 75%, whereas those of the Cornell and Sokolow-Lyon voltages were 50% and 12.5%, respectively. For predicting increased LVM at 80% specificity, the Cornell/QRS I I voltage ratio had a sensitivity of 56%, whereas the sensitivities of the Cornell and Sokolow-Lyon voltages were 36% and 22%, respectively. We conclude that, in constrast with the Sokolow-Lyon voltage, the new dimensionless Cornell/QRS I I voltage shows a sensitivity at a high specificity value at least as acceptable as that of the Cornell voltage for predicting borderline-high LVM in a population with a low prevalence of LVH.