The relation between serum potassium level and all-cause mortality was examined in a prospective study of 7,636 middle-aged British men followed for 11.5 years (1978-1991). Men being treated for hypertension had a significantly lower mean (+/- standard error) potassium level than men not in treatment (4.24 +/- 0.03 mmol/liter vs. 4.32 +/- 0.01 mmol/liter; p < 0.01). During the follow-up period of 11.5 years, after exclusion of the 374 men under antihypertensive treatment, there were 771 deaths from all causes in the remaining 7,262 men. A low potassium level (< 3.7 mmol/liter) was not associated with increased mortality. Elevated potassium levels > or = 5.2 mmol/liter were associated with a significant increase in mortality, particularly noncardiovascular deaths, even after adjustment for potentially confounding factors. However, serum potassium was strongly related to smoking, and the increased risk of mortality associated with elevated potassium was seen only among current smokers. In current smokers with raised potassium levels (> or = 5.2 mmol/liter) compared with smokers with levels under 5.2 mmol/liter, the relative risks of mortality were 1.7 (95% confidence interval (CI) 1.2-2.5) for deaths from all causes, 1.8 (95% CI 1.0-3.2) for all cancer deaths, and 2.5 (95% CI 1.1-5.6) for lung cancer deaths. In the 374 men receiving regular antihypertensive treatment, a low potassium level was not associated with excess mortality; those with raised potassium levels had excess risks for both cardiovascular and noncardiovascular deaths. The findings suggest that either raised potassium levels in association with smoking have an influence on the risk of death from noncardiovascular disease, particularly lung cancer, or a raised serum potassium level is a marker for some other risk factor associated with smoking. The prognostic and therapeutic implications of these observations warrant further exploration.