Objective This systematic review evaluated the diagnostic accuracy of Xpert MTB/RIF to detect tuberculous meningitis (TBM). Methods PubMed and five other databases were systematically searched through March 2019. All studies evaluating diagnostic accuracy of Xpert MTB/RIF on cerebrospinal fluid (CSF) samples were included. Reference standards were definitive or definite plus probable TBM. The quality of studies was assessed by the QUADAS‐2 tool. We performed bivariate random‐effects meta‐analysis and calculated summary diagnostic statistics. Results We identified 30 studies (n = 3972 participants), including 5 cohort studies and 25 cross‐sectional studies. Reference standards were definite TB (n = 28 studies) or definite plus probable TBM (n = 6 studies). The pooled Xpert MTB/RIF sensitivity was 85% (95% CI, 70–93%), and specificity was 98% (95% CI, 97–99%) with a negative likelihood ratio of 0.15 (95% CI, 0.04–0.27) for definite TBM. For probable TBM cases, pooled sensitivity was 81% (95% CI, 66–90%), and specificity was 99% (95% CI, 97–99%). For both reference standard types, meta‐analyses showed a C‐statistic area under the curve of 0.98. The QUADAS‐2 tool revealed low risk of bias as well as low concerns regarding applicability. Methodological heterogeneity was high among studies. Conclusions Xpert MTB/RIF showed high accuracy for TBM diagnosis, but a negative Xpert MTB/RIF test does not rule out TBM. Repeat Xpert testing may be necessary. In clinical practice, Xpert MTB/RIF adds speed and sensitivity when compared to classic TBM diagnostic methods or previous commercial nucleic acid amplification techniques. More studies and better strategies for rapidly confirming a diagnosis of TBM in children are urgently needed.