Background: Bedaquiline (BDQ) is critical for multidrug-resistant tuberculosis (MDR-TB) treatment, but emerging resistance threatens its efficacy. Detection of BDQ resistance relies on slow phenotypic drug-susceptibility testing (pDST), delaying treatment decisions. We determined the diagnostic accuracy of whole-genome sequencing (WGS) for predicting BDQ resistance against culture-based pDST. Methods: Mycobacterium tuberculosis isolates underwent BDQ pDST (1.0 µg/mL in MGIT) in a prospective study of patients with MDR-TB with additional resistance to fluoroquinolones or aminoglycosides in KwaZulu-Natal, South Africa (2018–2022). Isolates with single nucleotide polymorphisms showing a read depth of ≥10 and allele frequency of ≥0.7 were included. Isolates exhibiting phenotypic resistance underwent quantitative minimum inhibitory concentration (MIC) testing (ranging from 0.25 to 8.0 µg/mL) using the BACTEC MGIT 960 system. Diagnostic performance of WGS compared to MGIT-based pDST was assessed by calculating sensitivity, specificity, negative predictive value (NPV), and positive predictive value (PPV). Results: Analysis included 428 isolates: 88 phenotypically resistant isolates (87 with potential BDQ-resistance-associated variants (RAVs) and one wild-type) and 340 susceptible isolates (7 with potential RAVs and 333 wild-type). Presence of potential RAVs accurately predicted phenotypic resistance: sensitivity 98.9% (95% CI: 93.8–99.8), specificity 97.9% (95% CI: 95.8–99.0), PPV 92.6% (95% CI: 85.4–96.4), and NPV 99.7% (95% CI: 98.3–99.6). Among the 94 isolates with potential RAVs, we observed gene mutations in Rv0678 (88/94, 93.6%), pepQ (2/94, 2.1%), and Rv1979c (4/94, 4.3%). The median BDQ MIC for resistant isolates was 4.0 µg/mL [IQR:2.0–4.0 µg/mL]. Conclusions: Potential BDQ RAVs, predominantly in Rv0678, are strong predictors of BDQ phenotypic resistance, highlighting the potential use of genotypic catalogues to inform timely treatment decisions.