BACKGROUND: Despite causing high mortality worldwide, paediatric tuberculosis is often undiagnosed. We aimed to investigate optimal testing strategies for microbiological confirmation of tuberculosis in children younger than 15 years, including the yield in high-risk subgroups (eg, children younger than 5 years, with HIV, or with severe acute malnutrition [SAM]). METHODS: For this secondary analysis, we used data from RaPaed-TB, a multicentre diagnostic accuracy study evaluating novel diagnostic assays and testing approaches for tuberculosis in children recruited from five health-care centres in Malawi, Mozambique, South Africa, Tanzania, and India conducted between Jan 21, 2019, and June 30, 2021. Children were included if they were younger than 15 years and had signs or symptoms of pulmonary or extrapulmonary tuberculosis; they were excluded if they weighed less than 2 kg, had received three or more doses of anti-tuberculosis medication at time of enrolment, were in a condition deemed critical by the local investigator, or if they did not have at least one valid microbiological result. We collected tuberculosis-reference specimens via spontaneous sputum, induced sputum, gastric aspirate, and nasopharyngeal aspirates. Microbiological tests were Xpert MTB/RIF Ultra (hereafter referred to as Ultra), liquid culture, and Löwenstein-Jensen solid culture, which were followed by confirmatory testing for positive cultures. The main outcome of this secondary analysis was categorising children as having confirmed tuberculosis if culture or Ultra positive on any sample, unconfirmed tuberculosis if clinically diagnosed, and unlikely tuberculosis if neither of these applied. FINDINGS: Of 5313 children screened, 975 were enrolled, of whom 965 (99%) had at least one valid microbiological result. 444 (46%) of 965 had unlikely tuberculosis, 282 (29%) had unconfirmed tuberculosis, and 239 (25%) had confirmed tuberculosis. Median age was 5·0 years (IQR 1·8-9·0); 467 (48%) of 965 children were female and 498 (52%) were male. 155 (16%) of 965 children had HIV and 110 (11%) children had SAM. 196 (82%) of 239 children with microbiological detection tested positive on Ultra. 110 (46%) of 239 were confirmed by both Ultra and culture, 86 (36%) by Ultra alone, and 43 (18%) by culture alone. 'Trace' was the most common semiquantitative result (93 [40%] of 234). 481 (50%) of 965 children had only one specimen type collected, 99 (21%) of whom had M tuberculosis detected. 484 (50%) of 965 children had multiple specimens collected, 141 (29%) of whom were positive on at least one specimen type. Of the 102 children younger than 5 years with M tuberculosis detected, 80 (78%) tested positive on sputum. 64 (80%) of 80 children who tested positive on sputum were positive on sputum alone; 61 (95%) of 64 were positive on induced sputum, two (3%) of 64 were positive on spontaneous sputum, and one (2%) was positive on both. INTERPRETATION: High rates of microbiological confirmation of tuberculosis in children can be achieved via parallel sampling and concurrent testing procedures. Sample types and choice of test to be used sequentially should be considered when applying to groups such as children younger than 5 years, living with HIV, or with SAM. FUNDING: European and Developing Countries Clinical Trials Partnership programme, supported by the EU, the UK Medical Research Council, Swedish International Development Cooperation Agency, Bundesministerium für Bildung und Forschung, the German Center for Infection Research, and Beckman Coulter.