In 1996, the Hook Lake Wood Bison Recovery Project was initiated to establish a small, disease-free, captive, bison-breeding herd. Founders originated from wild bison herds in the Slave River Lowlands in northern Canada, which, like other bison herds in and around Wood Buffalo National Park, are endemically infected with bovine tuberculosis (caused by Mycobacterium bovis) and brucellosis (caused by Brucella abortus). After 9 yr of apparent disease freedom, tuberculosis was detected within the captive herd, leading to complete depopulation. This study examined the performance of antemortem tuberculosis diagnostic tests used during the project. Performances of the caudal-fold test, fluorescent polarization assay, multiantigen print immunoassay (MAPIA), and the rapid test (RT) were assessed by estimating sensitivity, specificity, positive predictive value, and negative predictive value for each test. Kappa values measuring agreement between tests were calculated. Overall, the tests did not differ with respect to sensitivities and specificities, which ranged from 50% to 92% and from 34% to 100%, respectively. The MAPIA tended to show high sensitivity, and there was significant agreement only between the MAPIA and RT. Serum collected from infected animals at slaughter produced highly variable results on the different assays, and one infected bison was negative on all antemortem tests. The results of this analysis suggest use of multiple antemortem tests in parallel, particularly those incorporating multiple antigens, to optimize sensitivity in detecting bovine tuberculosis in bison. However, as demonstrated in this herd, even a seemingly optimal antemortem testing regimen can fail to detect M. bovis–infected individuals.