When comparing several screening tests designed to detect the same disease, methodological problems arise in determining which is most accurate. We have previously demonstrated that receiver operating characteristic (ROC) methodology can provide a set of statistical procedures which allow for objective comparisons of screening tests. In this report, ROC methodology is brought to a different substantive area: the detection of clinical depression among treated alcoholics. In a sample of hospitalized alcoholics, we compared the accuracies of several commonly used screening tests for depressive disorders, as well as comparing each of these with a screening test for anxiety disorders. No test offered a statistically significant advantage over any other, and all did poorly in detecting clinically diagnosed major depression. The performance of the screening tests was worst when the non-depressed comparison group included subjects with remitted disorder, but was still poor when the comparison group did not include such potentially 'noisy' subjects. Factors contributing to the difficulty of screening for clinical depression are discussed, as well as suggestions for improvements in future screening efforts.