Data from a prospective study of 3,319 children ages 2 months to 5 years admitted with febrile illness to a Tanzanian district hospital were analyzed to determine the relationship of blood glucose and mortality. Hypoglycemia (blood sugar < 2.5 mmol/L and < 45 mg/dL) was found in 105 of 3,319 (3.2%) children at admission, and low-normal blood glucose (2.5-5 mmol/L and 45-90 mg/dL) was found in 773 of 3,319 (23.3%) children. Mortality was inversely related to admission blood sugar; compared with children with an admission blood glucose of > 5 mmol/L, the adjusted odds of dying were 3.3 (95% confidence interval = 2.1-5.2) and 9.8 (95% confidence interval = 5.1-19.0) among children with admission blood glucose 2.5-5 and < 2.5 mmol/L, respectively. Receiver operating characteristic (ROC) analysis suggested an optimal cutoff for admission blood sugar of < 5 mmol/L in predicting mortality (sensitivity = 57.7%, specificity = 75.2%). A cutoff for admission blood glucose of < 5 mmol/L represents a simple and clinically useful predictor of mortality in children admitted with severe febrile illness to hospital in resource-poor settings.