For the first time in thirty years the tuberculosis drug development pipeline may be on the verge of delivering ultra-short course chemotherapy but continued reliance on relapse rates in clinical trials and limited understanding ofthe phenomenon ofpersistence ofM tuberculosis during treatment have made the critical pathway of early clinical development uncertain. It is demonstrated here that serial sputum colony counting,prolonged throughout the first two months oftreatment and analysed using hierarchical non-linear regression techniques has the power to detect treatment differences between combination drug regimens at very modest sample sizes.