C-reactive protein (CRP) is used to discriminate common bacterial and viral infections, but its utility in tropical settings remains unknown. We performed a meta-analysis of studies performed in Asia and Africa. First, mean CRP levels for specific tropical infections were calculated. Thereafter, individual patient data (IPD) from patients with non-malarial undifferentiated fever (NMUF) who were tested for viral and bacterial pathogens were analysed, calculating separate cut-off values and their performance in classifying viral or bacterial disease. Mean CRP levels of 7307 patients from 13 countries were dengue 12.0 mg/l (standard error [SE] 2.7), chikungunya 41.0 mg/l (SE 19.5), influenza 15.9 mg/l (SE 6.3), Crimean-Congo haemorrhagic fever 9.7 mg/l (SE 4.7), Salmonella 61.9 mg/l (SE 5.4), Rickettsia 61.3 mg/l (SE 8.8), Coxiella burnetii 98.7 mg/l (SE 44.0) and Leptospira infections 113.8 mg/l (SE 23.1). IPD analysis of 1059 NMUF patients ≥5 y of age showed CRP <10 mg/l had 52% sensitivity (95% confidence interval [CI] 48 to 56) and 95% specificity (95% CI 93 to 97) to detect viral infections. CRP >40 mg/l had 74% sensitivity (95% CI 70 to 77) and 84% specificity (95% CI 81 to 87) to identify bacterial infections. Compared with routine care, the relative risk for incorrect classification was 0.64 (95% CI 0.55 to 0.75) and the number needed to test for one extra correctly classified case was 8 (95% CI 6 to 12). A two cut-off value CRP test may help clinicians to discriminate viral and bacterial aetiologies of NMUF in tropical areas.