OBJECTIVE--To investigate the prognostic value of frequently occurring slight variations in the ST segment for cardiovascular mortality in healthy subjects. DESIGN--Follow up study of mortality in relation to variations in ST segment level in a cohort over the 28 years from 1953 to 1981. A case-cohort sampling design was applied to limit the number of electrocardiograms that had to be coded by hand. SETTING--General health examination carried out in 1953 of civil servants in Amsterdam and assessment of subsequent mortality. SUBJECTS--Apparently healthy civil servants aged 40 to 65 years: 1583 men and 1508 women. MAIN OUTCOME MEASURES--Relative risk of variations in ST segment level for mortality from all causes, cardiovascular disease, and coronary heart disease. RESULTS--In men the multivariate relative risks of 15 year mortality from cardiovascular disease and coronary heart disease of slight ST elevation at 80 ms past the J point (compared with isoelectric ST segment) were 0.5 (95% confidence interval 0.3 to 0.9) and 0.4 (0.2 to 0.8), respectively. As expected, ST segment depression (greater than 0.25 mm) was associated with increased risk: 1.9 (1.1 to 3.0) and 2.2 (1.2 to 3.9), respectively. In women associations were weaker. The full 28 year period showed a similar pattern of somewhat weaker associations for men; among women, however, no predictive value was apparent. CONCLUSION--These results are empirical evidence for the intuitive opinion among doctors that a curved, upward sloping ST segment, resulting in slight ST elevation at 80 ms, indicates low risk compared with the isoelectric flat, stretched ST segment.