Background Pneumonia is one of the main causes of death ininfants in developing countries. The device of oxygen saturationmeasurement for detecting hypoxemia is limited in district hospi-tals.Objective The aim of our study was to find the best clinical pre-dictor for hypoxemia that could be used in Indonesia.Methods Between June 2001 until May 2002, the diagnostic testwas performed in 125 infants aged 2–12 month-old who sufferedfrom pneumonia. The oxygen saturation measured by pulse oxim-etry was used as the gold standard.Results The samples were divided into two groups, 52 infants withhypoxemia and 73 normal. The base characteristics of both groupswere not statistically different. The prevalence of hypoxemia was41.6%. The best single clinical predictor of hypoxemia was cyano-sis (the sensitivity 92%, specificity 86%, likelihood ratio 6.74, post-test probability 83%), as well as the combination of two clinicalpredictors i.e., cyanosis and nasal flaring. The best combination ofthree clinical predictors was cyanosis, nasal flaring, and refusal todrink (the sensitivity 92%, specificity 86%, likelihood ratio 6.74,post-test probability 81%).Conclusion The combination of cyanosis and nasal flaring isgood enough as a predictor to detect hypoxemia in area with nofacility of oxygen saturation measurement