A matched-pair analysis was conducted with data based on a prospective six-year follow-up of a random population sample to study the association between serum selenium and the risk of cancer. Case-control pairs were from a population, after exclusions, of 8,113 persons examined in 1972 from two counties in eastern Finland. Cases were 31- to 59-year-old men and women initially free of cancer. One control was matched to each case according to age, gender, daily tobacco consumption, and serum cholesterol concentration. The mean serum selenium of the 128 cases was 50.5 micrograms/liter and that of the controls was 54.3 micrograms/liter (p = 0.012 for difference). When the residual variation in tobacco consumption and serum cholesterol as well as that in four other possible confounders was allowed for in a multiple logistic model, serum selenium of less than 45 micrograms/liter was associated with a relative risk of cancer of 3.1 (95% confidence interval, 1.5-6.7, p less than 0.01). These data support the hypothesis that selenium deficiency increases the risk of certain cancers in middle-aged persons.