Characteristics of subjects with nonreproducible lung function tests (test failures) are described in two studies of occupational respiratory disease. According to current guidelines for spirometry, subjects with test failure are excluded from analyses of epidemiologic pulmonary function data. Among 415 Chinese cotton textile workers, the prevalence of byssinosis was 5.8% for subjects with repeatable tests and 13.3% for subjects with test failure. In regression analysis, the estimate of the association between cotton dust exposure and FEV1 decreased when subjects with test failure were excluded. In a second cohort of 378 asbestos-exposed machinists, the prevalence of chronic bronchitis was significantly greater among those with test failure. Considering only subjects with repeatable measurements, FEV1 was lower among textile workers with byssinosis and machinists with chronic bronchitis than among their asymptomatic coworkers. This suggests that, on average, subjects with poor test performance had lower (unobserved) FEV1 values, and therefore that the exclusion of subjects with test failure may cause selection bias.