Objective: The leukocyte count is frequently used to evaluate suspected bacterial infections but estimates of its test performance vary considerably. We evaluated its accuracy for the detection of serious bacterial infections in febrile children. Design: Prospective cohort study. Setting: Paediatric emergency department. Patients: Febrile 0-5-year-olds who had a leukocyte count on presentation. Outcome measures: Accuracy of total white blood cell and absolute neutrophil counts for the detection of urinary tract infection, bacteraemia, pneumonia and a combined ('any serious bacterial infection') category. Logistic regression models were fitted for each outcome. Reference standards were microbiological/radiological tests and clinical follow-up. Results: Serious bacterial infections were present in 714 (18.3%) of 3893 illness episodes. The area under the receiver operating characteristic curve for 'any serious bacterial infection' was 0.653 (95% CI 0.630 to 0.676) for the total white blood cell count and 0.638 (95% CI 0.615 to 0.662) for absolute neutrophil count. A white blood cell count threshold >15×10 9/L had a sensitivity of 47% (95% CI 43% to 50%), specificity 76% (95% CI 74% to 77%), positive likelihood ratio 1.93 (95% CI 1.75 to 2.13) and negative likelihood ratio 0.70 (95% CI 0.65 to 0.75). An absolute neutrophil count threshold >10×109/L had a sensitivity of 41% (95% CI 38% to 45%), specificity 78% (95% CI 76% to 79%), positive likelihood ratio 1.87 (95% CI 1.68 to 2.09) and negative likelihood ratio 0.75 (95% CI 0.71 to 0.80). Conclusions: The total white blood cell count and absolute neutrophil count are not sufficiently accurate triage tests for febrile children with suspected serious bacterial infection.