BACKGROUND: Pneumonia is the leading cause of death in children globally. Clinical algorithms remain suboptimal for distinguishing severe pneumonia from other causes of respiratory distress such as malaria or distinguishing bacterial pneumonia and pneumonia from others causes, such as viruses. Molecular tools could improve diagnosis and management. METHODS: We conducted a mass spectrometry-based proteomic study to identify and validate markers of severity in 390 Gambian children with pneumonia (n = 204) and age-, sex-, and neighborhood-matched controls (n = 186). Independent validation was conducted in 293 Kenyan children with respiratory distress (238 with pneumonia, 41 with Plasmodium falciparum malaria, and 14 with both). Predictive value was estimated by the area under the receiver operating characteristic curve (AUC). RESULTS: Lipocalin 2 (Lpc-2) was the best protein biomarker of severe pneumonia (AUC, 0.71 [95% confidence interval, .64-.79]) and highly predictive of bacteremia (78% [64%-92%]), pneumococcal bacteremia (84% [71%-98%]), and "probable bacterial etiology" (91% [84%-98%]). These results were validated in Kenyan children with severe malaria and respiratory distress who also met the World Health Organization definition of pneumonia. The combination of Lpc-2 and haptoglobin distinguished bacterial versus malaria origin of respiratory distress with high sensitivity and specificity in Gambian children (AUC, 99% [95% confidence interval, 99%-100%]) and Kenyan children (82% [74%-91%]). CONCLUSIONS: Lpc-2 and haptoglobin can help discriminate the etiology of clinically defined pneumonia and could be used to improve clinical management. These biomarkers should be further evaluated in prospective clinical studies.