Background: The diagnosis of pulmonary tuberculosis presents challenges in children because symptoms are non-specific, specimens are difficult to obtain, and cultures and smears of Mycobacterium tuberculosis are often negative. We assessed new diagnostic approaches for tuberculosis in children in a resource-poor country. Methods: Children with symptoms suggestive of pulmonary tuberculosis (cases) were enrolled from August, 2002, to January, 2007, at two hospitals in Lima, Peru. Age-matched and sex-matched healthy controls were enrolled from a low-income shanty town community in south Lima. Cases were grouped into moderate-risk and high-risk categories by Stegen-Toledo score. Two specimens of each type (gastric-aspirate, nasopharyngeal-aspirate, and stool specimens) taken from each case were examined for M tuberculosis by auramine smear microscopy, broth culture by microscopic-observation drug-susceptibility (MODS) technique, standard culture on Lowenstein-Jensen medium, and heminested IS6110 PCR. Specimens from controls consisted of one nasopharyngeal-aspirate and two stool samples, examined with the same techniques. This study is registered with ClinicalTrials.gov, number NCT00054769. Findings: 218 cases and 238 controls were enrolled. 22 (10%) cases had at least one positive M tuberculosis culture (from gastric aspirate in 22 cases, nasopharyngeal aspirate in 12 cases, and stool in four cases). Laboratory confirmation of tuberculosis was more frequent in cases at high risk for tuberculosis (21 [14·1%] of 149 cases with complete specimen collection were culture positive) than in cases at moderate risk for tuberculosis (one [1·6%] of 61). MODS was more sensitive than Lowenstein-Jensen culture, diagnosing 20 (90·9%) of 22 patients compared with 13 (59·1%) of 22 patients (p=0·015), and M tuberculosis isolation by MODS was faster than by Lowenstein-Jensen culture (mean 10 days, IQR 8-11, vs 25 days, 20-30; p=0·0001). All 22 culture-confirmed cases had at least one culture-positive gastric-aspirate specimen. M tuberculosis was isolated from the first gastric-aspirate specimen obtained in 16 (72·7%) of 22 cases, whereas in six (27·3%), only the second gastric-aspirate specimen was culture positive (37% greater yield by adding a second specimen). In cases at high risk for tuberculosis, positive results from one or both gastric-aspirate PCRs identified a subgroup with a 50% chance of having a positive culture (13 of 26 cases). Interpretation: Collection of duplicate gastric-aspirate specimens from high-risk children for MODS culture was the best available diagnostic test for pulmonary tuberculosis. PCR was insufficiently sensitive or specific for routine diagnostic use, but in high-risk children, duplicate gastric-aspirate PCR provided same-day identification of half of all culture-positive cases. Funding: National Institutes of Health. © 2010 Elsevier Ltd.