BackgroundAs access to Xpert expands in high TB-burden settings, its performance against clinically diagnosed TB as a reference standard provides important insight as the majority of childhood TB is bacteriologically unconfirmed. We aim to describe the characteristics and outcomes of children with presumptive TB and TB disease, and assess performance of Xpert under programmatic conditions against a clinical diagnosis of TB as a reference standard.MethodsRetrospective review of children evaluated for presumptive TB in Mbeya, Tanzania. Baseline characteristics were compared by TB disease status and, for patients diagnosed with TB, by TB confirmation status using Wilcoxon rank sum test for continuous variables and the Chi-square test for categorical variables. Sensitivity and specificity were calculated to assess the performance of Xpert, smear, and culture against clinical TB. Kappa statistics were calculated to assess agreement between Xpert and smear to culture.ResultsAmong children (N = 455) evaluated for presumptive TB, 70.3% (320/455) had Xpert and 62.8% (286/455) had culture performed on sputa. 34.5% (157/455) were diagnosed with TB: 80.3% (126/157) pulmonary TB, 13.4% (21/157) bacteriologically confirmed, 53.5% (84/157) HIV positive, and 48.4% (76/157) inpatients. Compared to the reference standard of clinical diagnosis, sensitivity of Xpert was 8% (95% CI 4–15), smear 6% (95% CI 3–12) and culture 16% (95% CI 9–24), and did not differ based on patient disposition, nutrition or HIV status.ConclusionDespite access to Xpert, the majority of children with presumptive TB were treated based on clinical diagnosis. Reflecting the reality of clinical practice in resource limited settings, new diagnostics such as Xpert serve as important adjunctive tests but will not obviate the need for astute clinicians and comprehensive diagnostic algorithms.Electronic supplementary materialThe online version of this article (doi:10.1186/s12879-017-2236-9) contains supplementary material, which is available to authorized users.