The Xpert MTB/RIF assay is a major advance for diagnosis of tuberculosis (TB) in high-burden countries but is limited in children by their difficulty to produce sputum. We investigated TB in sputum and stool from children with the aim of improving paediatric TB diagnosis. A prospective cohort of children with presumptive TB, provided two sputum or induced sputum at enrolment in a regional referral hospital in Uganda. Stool was collected from those started on TB treatment. All specimen were tested for Xpert MTB/RIF, mycobacteria growth indicator tube (MGIT), Lowenstein Jensen cultures and microscopy (except stool). We compared TB detection between age categories and assessed the performance of Xpert MTB/RIF in sputum and stool. Of the 392 children enrolled, 357 (91.1%) produced at least one sputum sample. Sputum culture yield was 13/357 (3.6%): 3/109 (2.6%), 3/89 (3.2%), 3/101 (2.6%) and 4/44 (8.2%) among children of 10 years, respectively (p = 0.599). Xpert MTB/RIF yield was 14/350 (4.0%): 3/104 (2.9%), 4/92 (4.3%), 3/88 (2.9%) and 4/50 (.0%), respectively (p = 0.283). Sensitivity and specificity of Xpert MTB/RIF in sputum against sputum culture were 90.9% (95% CI 58.7-99.8) and 99.1% (99.1-99.8). In stool, it was 55.6% (21.2-86.3) and 98.2% (98.2-100) against Xpert MTB/RIF and culture in sputum. Only a minority of children had microbiologically confirmed TB with a higher proportion in children above 10 years. Although sensitivity of Xpert MTB/RIF in stool was low, with good optimization, it might be a good alternative to sputum in children.