Background: Every year, at least one million children become ill with tuberculosis and around 200,000 children die. Xpert MTB/RIF and Xpert Ultra are World Health Organization (WHO)‐recommended rapid molecular tests that simultaneously detect tuberculosis and rifampicin resistance in adults and children with signs and symptoms of tuberculosis, at lower health system levels. To inform updated WHO guidelines on molecular assays, we performed a systematic review on the diagnostic accuracy of these tests in children presumed to have active tuberculosis. Objectives: Primary objectives: • To determine the diagnostic accuracy of Xpert MTB/RIF and Xpert Ultra for (a) pulmonary tuberculosis in children presumed to have tuberculosis; (b) tuberculous meningitis in children presumed to have tuberculosis; (c) lymph node tuberculosis in children presumed to have tuberculosis; and (d) rifampicin resistance in children presumed to have tuberculosis ‐ For tuberculosis detection, index tests were used as the initial test, replacing standard practice (i.e. smear microscopy or culture) ‐ For detection of rifampicin resistance, index tests replaced culture‐based drug susceptibility testing as the initial test Secondary objectives: • To compare the accuracy of Xpert MTB/RIF and Xpert Ultra for each of the four target conditions • To investigate potential sources of heterogeneity in accuracy estimates ‐ For tuberculosis detection, we considered age, disease severity, smear‐test status, HIV status, clinical setting, specimen type, high tuberculosis burden, and high tuberculosis/HIV burden ‐ For detection of rifampicin resistance, we considered multi‐drug‐resistant tuberculosis burden • To compare multiple Xpert MTB/RIF or Xpert Ultra results (repeated testing) with the initial Xpert MTB/RIF or Xpert Ultra result Search methods: We searched the Cochrane Infectious Diseases Group Specialized Register, MEDLINE, Embase, Science Citation Index, the Cumulative Index to Nursing and Allied Health Literature (CINAHL), Scopus, the WHO International Clinical Trials Registry Platform, ClinicalTrials.gov, and the International Standard Randomized Controlled Trials Number (ISRCTN) Registry up to 29 April 2019, without language restrictions. Selection criteria: Randomized trials, cross‐sectional trials, and cohort studies evaluating Xpert MTB/RIF or Xpert Ultra in HIV‐positive and HIV‐negative children younger than 15 years. Reference standards comprised culture or a composite reference standard for tuberculosis and drug susceptibility testing or MTBDRplus (molecular assay for detection of Mycobacterium tuberculosis and drug resistance) for rifampicin resistance. We included studies evaluating sputum, gastric aspirate, stool, nasopharyngeal or bronchial lavage specimens (pulmonary tuberculosis), cerebrospinal fluid (tuberculous meningitis), fine needle aspirates, or surgical biopsy tissue (lymph node tuberculosis). Data collection and analysis: Two review authors independently extracted data and assessed study quality using the Quality Assessment of Studies of Diagnostic Accuracy ‐ Revised (QUADAS‐2). For each target condition, we used the bivariate model to estimate pooled sensitivity and specificity with 95% confidence intervals (CIs). We stratified all analyses by type of reference standard. We assessed certainty of evidence using the GRADE approach. Main results: For pulmonary tuberculosis, 299 data sets (68,544 participants) were available for analysis; for tuberculous meningitis, 10 data sets (423 participants) were available; for lymph node tuberculosis, 10 data sets (318 participants) were available; and for rifampicin resistance, 14 data sets (326 participants) were available. Thirty‐nine studies (80%) took place in countries with high tuberculosis burden. Risk of bias was low except for the reference standard domain, for which risk of bias was unclear because many studies collected only one specimen for culture. Detection of pulmonary tuberculosis: For sputum specimens, Xpert MTB/RIF pooled sensitivity (95% CI) and specificity (95% CI) verified by culture were 64.6% (55.3% to 72.9%) (23 studies, 493 participants; moderate‐certainty evidence) and 99.0% (98.1% to 99.5%) (23 studies, 6119 participants; moderate‐certainty evidence). For other specimen types (nasopharyngeal aspirate, 4 studies; gastric aspirate, 14 studies; stool, 11 studies), Xpert MTB/RIF pooled sensitivity ranged between 45.7% and 73.0%, and pooled specificity ranged between 98.1% and 99.6%. For sputum specimens, Xpert Ultra pooled sensitivity (95% CI) and specificity (95% CI) verified by culture were 72.8% (64.7% to 79.6%) (3 studies, 136 participants; low‐certainty evidence) and 97.5% (95.8% to 98.5%) (3 studies, 551 participants; high‐certainty evidence). For nasopharyngeal specimens, Xpert Ultra sensitivity (95% CI) and specificity (95% CI) were 45.7% (28.9% to 63.3%) and 97.5% (93.7% to 99.3%) (1 study, 195 participants). For all specimen types, Xpert MTB/RIF and Xpert Ultra sensitivity were lower against a composite reference standard than against culture. Detection of tuberculous meningitis: For cerebrospinal fluid, Xpert MTB/RIF pooled sensitivity and specificity, verified by culture, were 54.0% (95% CI 27.8% to 78.2%) (6 studies, 28 participants; very low‐certainty evidence) and 93.8% (95% CI 84.5% to 97.6%) (6 studies, 213 participants; low‐certainty evidence). Detection of lymph node tuberculosis: For lymph node aspirates or biopsies, Xpert MTB/RIF pooled sensitivity and specificity, verified by culture, were 90.4% (95% CI 55.7% to 98.6%) (6 studies, 68 participants; very low‐certainty evidence) and 89.8% (95% CI 71.5% to 96.8%) (6 studies, 142 participants; low‐certainty evidence). Detection of rifampicin resistance: Xpert MTB/RIF pooled sensitivity and specificity were 90.0% (67.6% to 97.5%) (6 studies, 20 participants; low‐certainty evidence) and 98.3% (87.7% to 99.8%) (6 studies, 203 participants; moderate‐certainty evidence). Authors' conclusions: We found Xpert MTB/RIF sensitivity to vary by specimen type, with gastric aspirate specimens having the highest sensitivity followed by sputum and stool, and nasopharyngeal specimens the lowest; specificity in all specimens was > 98%. Compared with Xpert MTB/RIF, Xpert Ultra sensitivity in sputum was higher and specificity slightly lower. Xpert MTB/RIF was accurate for detection of rifampicin resistance. Xpert MTB/RIF was sensitive for diagnosing lymph node tuberculosis. For children with presumed tuberculous meningitis, treatment decisions should be based on the entirety of clinical information and treatment should not be withheld based solely on an Xpert MTB/RIF result. The small numbers of studies and participants, particularly for Xpert Ultra, limits our confidence in the precision of these estimates.