The aim of this study was to assess the role of high density lipoprotein (HDL)-cholesterol percentage and its relation to other variables assumed to be coronary risk factors. The 20-year follow-up involved 2,633 presumably healthy subjects, 1,308 men and 1,325 women, aged 25 to 69 years at the beginning of the study in 1964. Variables determined at entry examination included total cholesterol, HDL cholesterol, systolic and diastolic blood pressures, body mass index and cigarette smoking. During the 20-year period there were 242 fatal and nonfatal definite coronary events in men, and 108 in women. All variables differed in varying degrees between subjects who did and did not have a definite coronary event. Three multiple logistic regression models revealed that among these variables the most important was HDL-cholesterol percentage, which significantly improved the regression, even after total cholesterol was included with all other risk factors. With the first model, goodness-of-fit tests indicated that predicted values fit the observed values well, even after the first step, if HDL-cholesterol percentage is entered. The second and third models significantly fit the predicted values only after entering the HDL-cholesterol percentage. This means that although all the other variables, including total cholesterol, differed significantly between the affected and nonaffected groups, they were insufficient predictors alone. The results reveal that HDL-cholesterol percentage had the highest predictive value for risk of future coronary disease; it significantly improved the predictive capability of the logistic regression model, even after adjustment for all other mentioned variables. These findings indicate that the evaluation of coronary risk should not be made without determination of HDL-cholesterol percentage, especially in subjects with "normal" or marginally elevated total cholesterol.