Receiver operating characteristic (ROC) analysis is often used to assess the screening accuracy of continuous measures. Hanley and McNeil (Radiology 1983;148:839-843) developed a method for comparing ROC curves that is widely used in the literature. DeLong et al. (Biometrics 1988;44:837-845) subsequently developed an improved method; however, it has not become popular due to its perceived complexity. OBJECTIVE: This research compared the method of DeLong et al. to that of Hanley and McNeil. METHODS: The data were drawn from a sample of over 1,215 primary care patients participating in the Longitudinal Investigation of Depression Outcomes (LIDO) study. Three continuous measures (i.e., CES-D, MHI-5, and MCS-12) were compared in terms of their ability to detect diagnosis of major depression based on the CIDI. Values for area under the ROC curve (AUC) were computed for the three measures and compared statistically using the DeLong et al. and Hanley and McNeil methods. RESULTS: The two methods yielded entirely different statistical conclusions. Using the DeLong et al. Method, z statistics for difference between AUC values were 15-22% larger. The DeLong et al. method yielded confidence intervals that were 17-25% narrower than those derived using the method of Hanley and McNeil. The statistical efficiency of the Hanley and McNeil method relative to that of DeLong et al. ranged from 64-73%. CONCLUSIONS: The DeLong et al. method is more elegant and precise than that developed by Hanley and McNeil. The results of this investigation indicate that the former would require fewer patients than the latter to achieve a fixed level of statistical power.