OBJECTIVE: To evaluate the performance of Xpert Mycobacterium Tuberculosis/rifampicin (MTB/RIF) Ultra (Ultra) for diagnosis of childhood tuberculosis (TB) within public health systems. METHODS: In this cross-sectional study, children aged <15 years with presumptive pulmonary TB were consecutively recruited and evaluated for TB at tertiary-level hospitals in Benin, Mali, and Ghana. Bivariate random-effects models were used to determine the pooled sensitivity and specificity of Ultra against culture. We also estimated its diagnostic yield against a composite microbiological reference standard (cMRS) of positive culture or Ultra. RESULTS: Overall, 193 children were included in the analyses with a median (interquartile range) age of 4.0 (1.1-9.2) years, 88 (45.6%) were female, and 36 (18.7%) were HIV-positive. Thirty-one (16.1%) children had confirmed TB, 39 (20.2%) had unconfirmed TB, and 123 (63.7%) had unlikely TB. The pooled sensitivity and specificity of Ultra verified by culture were 55.0% (95% confidence interval [CI]: 28.0-79.0%) and 95.0% (95% CI: 88.0-98.0%), respectively. Against the cMRS, the diagnostic yield of Ultra and culture were 67.7% (95% CI: 48.6-83.3%) and 70.9% (95% CI: 51.9-85.8%), respectively. CONCLUSION: Ultra has suboptimal sensitivity in children with TB that were investigated under routine conditions in tertiary-level hospitals in three West African countries.