We compared the prognostic value of 12 electrocardiographic (ECG) variables in predicting risk of new-onset heart failure (HF) in a subgroup of 13,555 participants of the Atherosclerosis Risk in Communities (ARIC) study who were considered free of coronary heart disease at the onset of the study. Cox proportional hazards models were used to evaluate risk of HF for the highest decile of the distribution of each ECG variable (lowest decile for ST and T amplitudes in lead V(5)), with the remaining deciles as reference groups. Risk models were adjusted for demographic and clinical variables. In univariate Cox regression models, in men 11 and in women 8 of the 12 ECG variables were significant, strong predictors of risk of new-onset HF. Subsequently, 8 ECG variables with low mutual correlations were entered simultaneously into a multivariate Cox regression model. In men, large left ventricular mass by electrocardiogram, QT prolongation, and increased heart rate were the strongest independent predictors of new-onset HF, each with a twofold increased risk. Other independent predictors in men were ST depression in lead V(5), wide QRS/T angle, and old (silent) myocardial infarction, each with a >50% increased risk of incident HF. In women, QRS nondipolar voltage was associated with an 87% increased risk of incident HF, and other independent predictors, as in men, were wide QRS/T angle and increased heart rate. In conclusion, several ECG abnormalities are manifestations of evolving HF in men and women considered free of coronary heart disease.