Prediction of absolute coronary (CHD) risk is important for optimal targeting of antihypertensive and lipid-lowering treatment. Estimates of risk based on the Framingham population are widely used, but predictions from this population may not be valid for other populations. We have compared estimates of CHD risk in men free of vascular disease by four risk functions: the Framingham method, the Dundee method based on the UK Heart Disease Prevention Project, a method based on the British Regional Heart Study (RHS), and a risk function derived from the German PROCAM population. All variables required to calculate CHD risk by the four methods were collected prospectively for 206 consecutive hypertensive men recruited from general practices for an intervention trial, and new referrals to the Sheffield Hypertension Clinic. All men were free of vascular disease. The 206 men had mean (SD) age 58.7 (9.3) years; BP 154/ 91 (18/10) mmHg; cholesterol 6.0 (1.1) mmol/l; and HDL-C 1.1 (0.3) mmol/l. 18% were current smokers, 32% had parental death from CHD, 11% had diabetes and 2% had ECG-LVH. The annual risk of coronary events was calculated for each man using the Framingham, Dundee, RHS, and PROCAM methods. The methods were compared by the Bland-Altman method, Pearson correlations and paired t tests. The estimated annual CHD rates were for Framingham 2.3 (1.2)%; for Dundee 2.2 (1.4)%; for RHS 0.6 (0.5)%, and for PROCAM 2.3 (2.0). The RHS estimates were fourfold lower than the other methods (all p<0.00001). Bland-Altman plots showed no systematic error between Framingham, Dundee, and PROCAM methods, but a systematic error between these methods and the RHS. Correlations between the methods were: Framingham-PROCAM r=0.82, Framingham-Dundee r=0.68, Framingham-RHS r=0.55, Dundee-PROCAM r=0.66, PROCAM-RHS r=0.60. Dundee-RHS r=0.72. There was broad agreement between the Framingham, Dundee, and PROCAM methods as regards absolute risk and absence of systematic error. The correlations were relatively weak, with less agreement at very high CHD risk (>3%/year). The RHS method gave much lower estimates of CHD risk, but the high correlation with the Dundee method suggests a systematic (non random) difference which is as yet unexplained. The Dundee method has previously been validated against a second British population, and the Dundee, Framingham, and PROCAM estimates of risk are probably correct. If so, the RHS method is not useful in subjects free of vascular disease.