While studying knee osteoarthritis (OA) in the first Health and Nutrition Examination Survey, we unexpectedly found a protective association between smoking and OA. After adjustment for age, sex, and weight, smokers had a significantly lower rate of OA than did nonsmokers, and heavier smokers were less likely to have the disease than were light smokers. To test this association in a separate study and see if it was due to confounding factors, we looked at the Framingham Osteoarthritis Study, a study of elderly members of the Framingham Heart Study cohort. We evaluated whether the presence of knee OA in 1983-1985 was related to smoking status at the first Framingham examination, 36 years earlier. Subjects who had been smokers at examination 1 had a lower rate of OA (190 of 679, 28%) than did nonsmokers (276 of 736, 37.5%). In an analysis adjusted for age, sex, and weight, heavy smokers had a modestly lower risk of developing knee OA than did nonsmokers (relative risk 0.81). Also, the adjusted risk of severe OA was less in heavy smokers than in nonsmokers (relative risk 0.73). The negative association with OA persisted when we examined the average cigarette consumption over the first 10 years of the Framingham study. Furthermore, after controlling for age, sex, weight, knee injury history, sports activity history, physical activity level, coffee and alcohol consumption, and weight change after examination 1, and after modeling weight and age in a nonlinear manner, smoking remained a significant protector against later knee OA. It appears that smoking or some unidentified factor correlated with smoking modestly protects against the development of knee OA.