The aim of this study was to evaluate the Edinburgh Postnatal Depression Scale (EPDS) for screening and diagnosis of postpartum depression. Three months after delivery, EPDS was administered to 378 mothers from the 2004 Pelotas Birth Cohort Study, Rio Grande do Sul State, Brazil. Up to 15 days later, mothers were re-interviewed by mental health care professionals using a semi-structured interview based on ICD-10 (gold standard). We calculated the sensitivity and specificity of each cutoff point, and values were plotted as a receiver operator characteristic curve. The best cutoff point for screening postpartum depression was > 10, with 82.6% (75.3-89.9%) sensitivity and 65.4% (59.8-71.1%) specificity. For screening moderate and severe cases, the best cutoff point was > 11, with 83.8% (73.4-91.3%) sensitivity and 74.7% (69.4-79.5%) specificity. For diagnosis, EPDS was valid only for prevalence of postpartum depression in the 20-25% range, with 60% PPV for the > 13 cutoff point (59.5% sensitivity; 88.4% specificity). The specificities and PPVs for all cutoff points were below those reported by other authors. Small numbers and the calculation of PPV in samples with overrepresentation of cases in the majority of studies appear to account for these differences.