Objective: To examine the relationship between body mass index (BMI) at age 70, weight change between age 70 and 75, and 15 y mortality. Design: Cohort study of 70-y-olds. Setting: Geriatric Medicine Department, Go Èteborg University, Sweden. Subjects: A total of 2628 (1225 males and 1403 females) 70-y-olds examined in 1971 ± 1981 in Gothenburg, Sweden. Results: The relative risks (RRs) for 15 y mortality were highest in the lowest BMI quintiles of males 1.20 (95% CI 0.96 ± 1.51) and females 1.49 (95% CI 1.14 ± 1.96). In non-smoking males, no signi®cant differences were observed across the quintiles for 5, 10 and 15 y mortality. In non-smoking females, the highest RR (1.58, 95% CI 1.15 ± 2.16) for 15 y mortality was in the lowest quintile. After exclusion of ®rst 5 y death, no excess risks were found in males for following 5 and 10 y mortality across the quintiles. In females, a U-shaped relation was observed after such exclusions. BMI ranges with lowest 15 y mortality were 27 ± 29 and 25 ± 27 kgam 2 in nonsmoking males and females, respectively. A weight loss of ! 10% between age 70 and 75 meant a signi®cantly higher risk for subsequent 5 and 10 y mortality in both sexes relative to individuals with`stable' weights. Conclusion: Low BMI and weight loss are risk factors for mortality in the elderly and smoking habits did not signi®cantly modify that relationship. The BMI ranges with lowest risks for 15 y mortality are relatively higher in elderly. Exclusion of early deaths from the analysis modi®ed the weight ± mortality relationship in elderly males but not in females.