To measure the performance of the current WHO algorithm in identifying children at higher risk of death, children aged 2-59 months who presented with cough and/or difficult breathing and were admitted into the paediatric hospital of Bangui (Central African Republic) during a 1-year period (1996/97) were investigated. Among children with subcostal indrawing, mortality and severity of oxygen desaturation were identical whether or not they also had tachypnoea. Among children with a 'severe pneumonia', those who also fulfilled the 'very severe disease' definition had a higher risk of death (31/132, 23.5%) than those who did not (12/106, 11.3%, P = 0.02). However, this 'very severe disease' definition did not predict death when used in children who did not have severe pneumonia. To identify variables that would better predict death, combinations of symptoms and signs were examined among the subgroup of children with indrawing. Nine combinations had both a sensitivity and specificity over 60%. 'Grunting and/or nasal flaring' had a sensitivity of 72% and a specificity of 66% in predicting death, and might be easier to use by primary health care personnel than other combinations. A new algorithm is proposed for the management of children aged 2-59 months presenting with cough and/or difficult breathing. The definition of pneumonia would be unchanged (tachypnoea). Severe pneumonia would remain defined on indrawing regardless of respiratory rate, except that indrawing should be lower chest wall and/or intercostal. In health facilities where intravenous antibiotics, chloramphenicol and/or oxygen are available, entry into a 'very severe pneumonia' category would be based on 'grunting and/or nasal flaring' among children with indrawing. In our study population, the mortality rates in the categories based on these definitions were 0.8% (1/127) in children with no pneumonia, 0.9% (1/116) in children with pneumonia, 7.7% (12/156) in children with severe pneumonia and 31.1% (33/106) in children with very severe pneumonia.