Body mass index (weight/height2) was studied for its prediction of lung cancer in Finnish men examined by the Social Insurance Institution's Mobile Clinic and followed up by the Finnish Cancer Registry. Among 25,994 participants aged 20 to 75 years and free from cancer at the start of the study, 504 lung-cancer cases were diagnosed during a maximum follow-up of 19 years. There was a significant inverse gradient between body mass index and the incidence of lung cancer. This association was not due to confounding by age, smoking, social class, self-perceived general health, history of stress symptoms, chronic cough or chest X-ray findings, although these factors correlated with body mass index and were also significantly predictive of lung cancer. After adjustment for these factors, the relative risks of lung cancer were 1.0, 1.4, 1.5 and 1.8 (p for trend < 0.001) from the highest to the lowest quartiles of body mass index. The association was noted among current smokers, but it was most obvious among men who had never smoked, the relative risk of lung cancer between the lowest and highest quartiles of body mass index being 7.6 (95% confidence interval = 2.0 to 29.4). The inverse gradient persisted throughout the entire follow-up period, and was most distinct after the first 10 years of follow-up, the relative risks for the quartiles of body mass index being 1.0, 1.9, 2.3 and 2.6 (p < 0,001). The present study shows that the thinner the man, the more liable he is to develop lung cancer. Leanness, or some factor or cluster of factors closely correlated to it, may play an aetiological role in lung cancer.