The relationship among body mass index (BMI, kg/m2), smoking status, and overall mortality remains controversial. To assess this relationship in a representative sample of older women, we used data from the Panel Study of Income Dynamics (PSID). The PSID (begun in 1968) is a prospective longitudinal cohort study designed to examine economic and demographic behavior. Respondents were 1355 women age > or = 50 when they initially completed the Self-Administered Health Questionnaire in 1990. Data collected included self-reported height and weight, years of completed education, smoking status (never versus ever), and responses to four health-related questions (e.g., retired due to ill health, difficulty eating). Respondents were followed, including the date of death if respondent died, through 1994. Cox proportional hazard regression revealed a U-shaped relationship irrespective of whether smoking was included in the model. The base of the curve was fairly wide, suggesting that a broad range of BMI is well tolerated by older women. The minimum mortality (estimated from fitted proportional hazard models) for both the smoking and nonsmoking models occurred at a BMI of approximately 34. When interactions between smoking status and BMI terms were added to the model, the interactions were not jointly significant (p = 0.071). Moreover an exploratory plot of the BMI-mortality curve among never smokers (n = 800) revealed a curve that moved away from being U-shaped toward being more monotonically decreasing. It is concluded that these data suggest that there is no evidence that the U-shaped BMI-mortality relationship observed is caused by confounding by smoking status.