The association of pulmonary function (as percent of predicted forced expiratory volume in 1 second [FEV1]) with total and cause-specific mortality over 15 to 18 years was investigated in a large cohort (5924) of prospectively followed Japanese-American men. Among those who never smoked, pulmonary function was found not to be significantly predictive of total mortality in a multivariate model in which adjustment for variables that might confound the results was made. Among past and current smokers, highly significant associations were found (P < 0.0001). The positive relationship of pulmonary function to mortality in smokers was so strong that it overshadowed these differences in nonsmokers in a model including all smoking groups combined, even after adjusting for smoking. A smoking-pulmonary function interaction term added to this model was statistically significant (P < 0.003). This illustrates the need for attention to the potential for complex interactions between biologic variables when carrying out multivariate statistical analysis. Findings for cardiovascular and noncardiovascular mortality were similar. This analysis indicates that while pulmonary function is associated with subsequent mortality, the relationship is significantly associated with smoking history.