The criterion validity of the Center for Epidemiological Studies Depression (CES-D) scale was evaluated among 101 orthopaedic and 50 neurological inpatients of a rehabilitation centre. The structured clinical interview for the DSM-IV (SCID-I) was used as criterion. Sensitivity of the CES-D to current major depressive disorder was 100% in both groups, while specificity was 57% (95% confidence interval = 48-67, likelihood ratio = +2.34) in orthopaedic patients and 36% (95% confidence interval = 23-49, likelihood ratio = +1.56) in neurological patients. Positive predictive value of the CES-D for current major depressive disorder was 24% (95% confidence interval = 10-32, likelihood ratio = +2.34) in orthopaedic patients and 31% in neurological patients (95% confidence interval = 18-43, likelihood ratio = +1.56), while negative predictive value was 100% in both groups. When a broader range of depressive disorders was considered, sensitivity dropped to 89% (95% confidence interval = 83-95, likelihood ratio = +3.52) and 96% (95% confidence interval = 91-100, likelihood ratio = +2.21) while specificity increased to 75% (95% confidence interval = 66-83, likelihood ratio = +3.52) and 57% (95% confidence interval = 43-70, likelihood ratio = +2.21), respectively.