The Bergen Clinical Blood Pressure Study in Norway was used to examine the relationship between body mass index (BMI (kg/m(2))) and total mortality in different age segments. Of 6,811 invited subjects, 5,653 (84%) participated in the study (1965-1971) and 4,520 (66%) died during 182,798 person-years of follow-up (1965-2007). Mean age at baseline was 47.5 years; range 22-75 years. BMI (kg/m(2)) was calculated from standardized measurements of body height and weight and divided into four groups (<22.0, 22.0-24.9, 25.0-27.9, > or =28.0). The 20 years cumulative risk of death related to baseline BMI was U-shaped in the elderly (aged 65-75 years), whereas the pattern was more linear in the youngest age group (20-44 years). In contrast to the younger age groups, the highest mortality in the elderly was in the lower BMI range (<22.0 kg/m(2)) (adjusted Cox proportional Hazard Ratio 1.39, 95% Confidence Interval 1.10, 1.75) compared to the BMI reference group (22.0-24.9 kg/m(2)). This pattern persisted after 72 months of early follow-up exclusion and it was robust to adjustments for a wide range of possible confounders including gender, history of cardiovascular disease, respiratory disease or hypertension, smoking habits, physical activity, socioeconomic status, physical appearance and other anthropometric measures. The study shows that a low BMI is an appreciable independent risk factor of total mortality in the elderly, and not a result of subclinical disease or confounding factors such as current or previous smoking. Awareness of this issue ought to be emphasized in advice, care and treatment of elderly subjects.