Bacteriological confirmation of Mycobacterium tuberculosis is achieved in the minority of young children with tuberculosis (TB), since specimen collection is resource intensive and respiratory secretions are mostly paucibacillary, leading to limited sensitivity of available diagnostic tests. Although molecular tests are increasingly available globally, mycobacterial culture remains the gold standard for diagnosis and determination of drug susceptibility and is more sensitive than molecular methods for paucibacillary TB. We evaluated stool culture as an alternative to respiratory specimens for the diagnosis of suspected intrathoracic TB in a subgroup of 188 children (median age, 14.4 months; 15.4% HIV infected) enrolled in a TB diagnostic study at two local hospitals in Cape Town, South Africa. One stool culture was compared to overall bacteriological confirmation by stool Xpert and by Xpert and culture of multiple respiratory specimens. After decontamination/digestion with NALC (N-acetyl-l-cysteine)-NaOH (1.25%), concentrated fluorescent smear microscopy, Xpert MTB/RIF, and liquid culture were completed for all specimens. Culture contamination of stool specimens was high at 41.5%. Seven of 90 (7.8%) children initiating TB treatment were stool culture positive for M. tuberculosis. Excluding contaminated cultures, the sensitivity of stool culture versus confirmed TB was 6/25 (24.0%; 95% confidence interval [CI] = 9.4 to 45.1%). In addition, stool culture detected TB in 1/93 (1.1%) children with “unconfirmed TB.” Testing the same stool by Xpert increased sensitivity to 33.3% (95% CI = 18.0 to 51.8%). In conclusion, stool culture had low sensitivity for M. tuberculosis detection in children with intrathoracic TB. Reducing culture contamination through improved laboratory protocols may enable more reliable estimates of its diagnostic utility.