The lower limit of normal (LLN) in pulmonary diffusing capacity has been determined by using the lower 5th percentile or 95% confidence interval (CI) obtained from a normal reference population. We aimed to test clinical cut-off values (1-99% CI) and to determine the LLN in a population with interstitial lung disease and in healthy subjects, using receiver operating characteristic (ROC) curve analysis. The best cut-off value of the LLN for differentiating subjects with pathologic proven interstitial lung disease (ILD) from healthy subjects was not the 95% CI, but the 90% CI, which gave 95% sensitivity and 100% specificity. Thus, the 90% CI or the lower 10th percentile for the cut-off value of the LLN might be better than the 95% CI or the lower 5th percentile in differentiating subjects with pathologic proven ILD from healthy subjects.