Xpert MTB/RIF assay, a real-time PCR assay designed to detect Mycobacterium tuberculosis, has proven sensitive and specific when performed on respiratory samples in a high prevalence setting. However, it was suggested as less accurate in a low-incidence environment. We evaluated the accuracy of the Xpert for the diagnosis of tuberculosis (TB) on pulmonary and extrapulmonary samples in Geneva (Switzerland), where the prevalence of active TB is very low. From March 2009 to February 2013, the Xpert was performed on clinical samples. All specimens were also processed using auramine, AFB staining, and mycobacterial culture with both solid and liquid media. The accuracy of both microscopy and Xpert was determined retrospectively using cultures as the reference method. A total of 732 clinical specimens were processed with the Xpert. The Xpert had a high specificity (97.5%; 95% confidence interval (CI), 95.8-98.5%) and revealed much more sensitive (82.7%; 95% CI, 74.1-89.0%) than microscopy (55.5%; 95% CI, 45.7-64.8%) for the diagnosis of TB, with a high negative predictive value (96.8%; 95% CI, 95.0-98.0%). The advantage of PCR over microscopy was even more pronounced for extrapulmonary specimens (sensitivity of 70% (95% CI, 50.4-84.6%) compared with 23.3% (95% CI, 10.6-42.7%)). Despite the low prevalence of TB in Switzerland, results performance for respiratory samples was similar to that reported in high prevalence countries. The high negative predictive value is clinically helpful in our setting, where pulmonary TB needs to be reasonably ruled out. When considering extrapulmonary samples, microscopy performed poorly compared with Xpert. This study shows that the Xpert remains accurate and useful in a low-incidence setting.