Global initiatives have been launched to develop improved tuberculosis chemotherapy. New drugs and potential treatment-shortening regimens require careful assessment in clinical trials, but existing markers of treatment outcome—clinical cure and relapse—require prolonged follow-up of patients. There is, therefore, a need to find alternative biomarkers or surrogate endpoints predictive of response. Effective treatment requires drugs with sterilising activity to produce clinical cure without relapse, and thus a useful biomarker for a drug under trial must predict the likelihood of relapse. We explore the strengths and weaknesses of existing biomarkers, which assess either host response or mycobacterial load. Change in mycobacterial burden is likely to be the best indicator of treatment outcome, but the optimum study techniques remain undefined. Finally, we propose methods to assess candidate markers, and how these candidate markers could be implemented in future clinical trials.