WHO-endorsed phenotypic drug susceptibility testing (DST) methods for Mycobacterium tuberculosis are assumed to be the gold standard for identifying rifampin (RMP) resistance. However, previous results indicated that low-level, yet probably clinically relevant, RMP resistance linked to specific rpoB mutations is easily missed by some growth-based methods. We aimed to compare the level of resistance detected on Löwenstein-Jensen (LJ) medium with resistance detected by the Bactec MGIT 960 automated DST (MGIT-DST) system for various rpoB mutants. Full agreement between LJ and MGIT-DST was observed for mutations located at codons 513 (Lys or Pro) and 531 (Leu, Trp), which were always resistant by both methods. For mutations 511Pro, 516Tyr, 533Pro, 572Phe, and several 526 mutations, LJ and MGIT results were highly discordant, with MGIT-DST failing to give a result or declaring the strains susceptible. Our data show that phenotypic RMP resistance testing of M. tuberculosis is not a binary phenomenon for some rpoB mutations and that the widely used automated MGIT 960 system is prone to miss some RMP resistance-conferring mutations, while careful DST on LJ missed hardly any. Given the association of these mutations with poor clinical outcome, our findings suggest that the gold standard for rifampin resistance should be reconsidered, in order to address the present confusion caused by discrepancies between phenotypic and genotypic results. The impacts of these mutations will depend on the frequency of their occurrence, which may vary from one setting to another.