Objective: We hypothesized that low serum selenium was associated with anemia in humans. Subjects: A total of 2092 adults aged 65 and older, in the third National Nutrition Examination Survey, Phase 2 (1991-1994) (NHANES III). Methods: Examination of the relationship between serum selenium and hematological indices in NHANES III. Results: Anemia, defined by World Health Organization criteria, was present in 12.9%. Mean serum selenium among non-anemic and anemic adults was 1.60 and 1.51 mmol l À1 (P ¼ 0.0003). The prevalence of anemia among adults in the lowest to highest quartiles of serum selenium was 18.3, 9.5, 9.7 and 6.9%, respectively (P ¼ 0.0005). The proportion of adults in the lowest quartile of selenium among those who were non-anemic or who had anemia due to nutritional causes, chronic inflammation, renal disease or unexplained anemia was 9.9, 27.5, 17.5, 24.0 and 15.4%, respectively. An increase in log e selenium was associated with a reduced risk of anemia (odds ratio per one standard deviation increase 0.75, 95% confidence interval 0.58-0.97, P ¼ 0.03), adjusting for age, race, education, body mass index and chronic diseases. Conclusion: Low serum selenium is independently associated with anemia among older men and women in the United States.