Risk factors for baseline bedaquiline (BDQ) resistance, amplification during treatment, and correlations with treatment outcomes are not fully understood. This cohort included Armenian patients with multidrug-resistant TB predominantly fluoroquinolone-resistant enrolled between 2013 and 2015 in a BDQ compassionate use program. BDQ resistance at baseline and during treatment was assessed using MGIT (pDST), minimal inhibitory concentration in 7H11 (MIC), and whole-genome sequencing. Risk factors, such as treatment effectiveness or stage of the disease, were analyzed for association with baseline BDQ resistance, acquired BDQ resistance, and treatment outcome. Among 39 patients, baseline BDQ resistance was 6% (2/33) by pDST and 7% (2/29) by MIC. All four baseline isolates with an mutation were phenotypically resistant. During treatment, 48% of the patients acquired BDQ resistance by pDST, and 52% acquired mutations at various frequencies (97% in ). None of the factors significantly contributed to baseline or acquired BDQ resistance. Unfavorable treatment outcome (41%) was more frequent in the presence of acquired mutations [odds ratio (OR) 132, 95% confidence interval (CI) 7.43, 2375], phenotypic BDQ resistance (OR 176, 95% CI 6.48, 2423), or MIC increase above or below the critical concentration (both OR 84.3, 95% CI 2.93, 2423) during treatment. For these highly treatment-experienced patients, low baseline prevalence but high incidence of acquired BDQ resistance was observed. Acquisition of mutations in BDQ candidate resistance genes, regardless of their frequency, or increased MICs during treatment, even below the critical concentration, should be seen as a warning sign of resistance amplification and increased risk of unfavorable treatment outcome.