A hospital-based follow-up study was conducted between 1986 and 1988 at Lwiro (South Kivu Province, Zaire). Of 1,129 children in the study, three of four were severely malnourished, and 17.4% died. This study analyzes the mortality in hospital; its objectives are to evaluate the prognostic power of edema and anthropometric and biologic indicators and to seek indices that perform better. Receiver operating characteristic curves were established for each parameter under study and for each index constructed. Areas under receiver operating characteristic curves were highest for biologic indicators, and simple indices, obtained by counting the number of risk factors present, performed best. In the absence of biologic parameters, the authors suggest classifying children as at risk of dying when they present with edema and/or with arm circumference of less than 115 mm. When biologic measurements are possible, in addition to edema and arm circumference, the authors suggest taking serum albumin and transthyretin into account. For serum albumin and transthyretin, mortality risk is defined in terms of values of less than 16 g/liter and 6.5 mg/dl, respectively. Children will be classified as at risk of dying when they present with at least two of the four risk factors. The resulting diagnostic test has a high sensitivity (91.2%) and positive and negative predictive values of 40.8% and 97.9%, respectively.