We aimed to investigate the diagnostic value of applying cut-off levels of inflammatory markers and to develop a prediction model for differentiation between bacterial and viral infections in paediatric community-acquired pneumonia based on C-reactive protein (CRP), neutrophil and white cell counts (WCC). Among 401 children, those with bacterial pneumonia were older than those with viral pneumonia (P < 0.001). Compared to viral, bacterial infections had a higher median CRP level (P < 0.001), whereas WCC and neutrophil count were not different. Bacterial infections were associated with higher CRP >80 mg/L than viral infections (P = 0.001), but levels <20 mg/L were not discriminatory (P = 0.254). ROC curve of the model for differentiating bacterial from viral pneumonia based on age, CRP and neutrophil count produced area under the curve of 0.894 with 75.7% sensitivity and 89.4% specificity. This aetiological discriminant prediction model is a potentially useful tool in clinical management and epidemiological studies of paediatric pneumonia.