Retinol and retinol-binding protein levels were measured in sera previously obtained, and stored in the frozen state, at multiphasic health checkups from 151 persons subsequently found to have lung cancer (cases) and 302 persons who remained free of cancer (controls). Two controls were matched to each case for sex, skin color, age, date of multiphasic health checkup, and aspects of the smoking habit. Mean levels in cases and controls were, respectively, retinol: 82.17 and 82.37 micrograms/dl (p = 0.93), and retinol-binding protein: 6.04 and 6.00 mg/dl (p = 0.81). Mean differences between cases and controls were, retinol: 0.195 micrograms/dl with 95% confidence limits, -3.91 and 4.30 micrograms/dl; retinol-binding protein: -0.033 mg/dl with 95% confidence limits, -0.31 and 0.24 mg/dl. No significant trend in relative risk of lung cancer was observed when the retinol or retinol-binding protein distribution was divided into quintiles. No significant associations were observed in subgroups based on age, sex, histologic type of cancer, cigarette consumption, or interval between blood drawing and cancer diagnosis. In this large study, retinol and retinol-binding protein levels were not useful in predicting the subsequent development of lung cancer.