OBJECTIVE--To estimate the size of the association between serum concentration of low density lipoprotein cholesterol and mortality from ischaemic heart disease. DESIGN--Prospective study of total serum cholesterol concentration and mortality from ischaemic heart disease in 21,515 men (538 deaths) and study of total cholesterol concentration measured on two occasions an average of three years apart in 5696 men in whom low density lipoprotein cholesterol concentration was also measured on the second occasion. SUBJECTS--Men who attended the medical centre of the British United Provident Association (BUPA) in London between 1975 and 1982. MAIN OUTCOME MEASURE--The difference in mortality from ischaemic heart disease for a 0.6 mmol/l difference in concentration of low density lipoprotein cholesterol after adjustment for, firstly, regression dilution bias, which arises from the random fluctuation of serum cholesterol concentration in people over time, and, secondly, the surrogate dilution effect, which arises because differences in total cholesterol concentration between people reflect smaller differences in low density lipoprotein cholesterol concentration. RESULTS--The observed difference in mortality from ischaemic heart disease associated with a difference of 0.6 mmol/l in total serum cholesterol concentration was 17% but increased to 24% after correction for the regression dilution bias and to 27% (95% confidence interval 21% to 33%) after adjustment for both sources of underestimation, which provides an estimate of the difference in mortality for a true difference of 0.6 mmol/l in low density lipoprotein cholesterol concentration. The association was greater at younger ages. The estimated decrease in mortality from all causes was 6% before and 10% (1% to 17%) after adjustment for the two sources of underestimation. There was no excess mortality from any cause associated with low cholesterol concentration. CONCLUSIONS--The association between serum cholesterol concentration and ischaemic heart disease is materially stronger than directly inferred from prospective studies. This has important implications for the health benefit of achieving low cholesterol concentrations.