A validity study of the Brazilian version of the 15-item Short Psychiatric Evaluation Schedule (SPES), included in the mental health assessment of Older Americans Resources and Services (OARS), designed to detect psychiatric disorders in the elderly, against the "caseness" criterion suggested by Cooper and Schwarz was carried out with a community sample, as part of a survey to study health and living conditions of the elderly in a large urban center of a developing country, São Paulo, Brazil. The screening questionnaire was completed by 292 subjects, and 91 were selected for the psychiatric interview. The validity coefficients were as follows: sensitivity 61%, specificity 89%, positive predictive value 66%, negative predictive value 87% and misclassification rate 18%. A discriminant analysis using a stepwise procedure was then applied to select the best item discriminators of the screening questionnaire. The best set of discrimination comprised six items leading to the following validity coefficients: sensitivity 82%, specificity 77%, positive predictive value 58%, negative predictive value 92% and misclassification rate 21%. The possible factors related to false positive and false negative responses on the screening are discussed.