Although the evidence base regarding the use of the Xpert MTB/RIF assay for diagnosis of pulmonary tuberculosis (TB) when testing respiratory samples is well established, the evidence base for its diagnostic accuracy for extrapulmonary and sputum-scarce pulmonary TB when testing non-respiratory samples is less clearly defined. A systematic literature search of 7 electronic databases (Medline, EMBASE, ISI Web of Science, BIOSIS, Global Health Database, Scopus and Cochrane Database) was conducted to identify studies of the diagnostic accuracy of the Xpert assay when testing non-respiratory samples compared with a culture-based reference standard. Data were extracted and study quality was assessed using the QUADAS-2 tool. Sensitivities and specificities were calculated on a per-sample basis, stratified by sample type and smear microscopy status and summarised using forest plots. Pooled estimates were calculated for groups with sufficient data. Twenty-seven studies with a total of 6,026 non-respiratory samples were included. Among the 23 studies comparing Xpert and culture done on the same samples, sensitivity was very heterogeneous with a median sensitivity of 0.83 (IQR, 0.68-0.94) whereas specificities were typically very high (median, 0.98; IQR, 0.89-1.00). The pooled summary estimates of sensitivity when testing smear-positive and smear-negative samples were 0.95 (95% CI 0.91-1.00) and 0.69 (95% CI 0.60-0.80), respectively. Pooled summary estimates of sensitivity varied substantially between sample types: lymph node tissue, 0.96 (95% CI, 0.72-0.99); tissue samples of all types, 0.88 (95% CI, 0.76-0.94); pleural fluid, 0.34 (95% CI, 0.24-0.44); gastric aspirates for diagnosis of sputum-scarce pulmonary TB, 0.78 (IQR, 0.68 - 0.85). Median sensitivities when testing cerebrospinal fluid and non-pleural serous fluid samples were 0.85 (IQR, 0.75-1.00) and 0.67 (IQR, 0.00-1.00), respectively. Xpert detects with high specificity the vast majority of EPTB cases with smear-positive non-respiratory samples and approximately two-thirds of those with smear-negative samples. Xpert is a useful rule-in diagnostic test for EPTB, especially when testing cerebrospinal fluid and tissue samples. In addition, it has a high sensitivity for detecting pulmonary TB when using gastric aspirate samples. These findings support recent WHO guidelines regarding the use of Xpert for TB diagnosis from non-respiratory samples.