Identifying a blood circulating cellular biomarker that can be used to assess severity of disease and predict the time to culture conversion (TCC) in patients with multidrug resistant tuberculosis (MDR-TB) would facilitate monitoring response to treatment and may be of value in the design of future drug trials. We report on the frequency of blood Ki67+HLA-DR− CD4+ T regulatory (Treg) cells in predicting microbiological outcome before initiating second-line treatment for MDR-TB. Fifty-one patients with MDR-TB were enrolled and followed over 18 months; a subset of patients was sputum culture (SC) negative at baseline (n = 9). SC positive patients were divided into two groups, based on median TCC: rapid responders (≤71 days TCC; n = 21) and slow responders (>71 days TCC; n = 21). Whole blood at baseline, months 2 and 6 was stimulated with M tuberculosis (Mtb) antigens and Treg cells were then identified as CD3+CD4+CD25hiFoxP3+CD127−CD69− and further delineated as Ki67+HLA-DR− Treg. The frequency of these cells was significantly enlarged at baseline in SC positive relative to SC negative and smear positive relative to smear negative patients and in those with lung cavitation. This difference was further supported by unsupervised hierarchical clustering showing a significant grouping at baseline of total and early differentiated memory Treg cells in slow responders. Conversely, there was a clustering of a lower proportion of Treg cells and activated IFNγ-expressing T cells at baseline in the rapid responders. Examining changes over time revealed a more gradual reduction of Treg cells in slow responders relative to rapid responders to treatment. Receiver operating curve analysis showed that baseline Mtb-stimulated Ki67+HLA-DR− Treg cells could predict the TCC of MDR-TB treatment response with 81.2% sensitivity and 85% specificity (AUC of 0.87, p < 0.0001), but this was not the case after 2 months of treatment. In conclusion, our data show that the frequency of a highly defined Mtb-stimulated blood Treg cell population at baseline can discriminate MDR-TB disease severity and predict time to culture clearance.