The urgent need for new anti-tuberculosis drugs raises the question of the design, conduct and analysis of the trials that will be required for licensing purposes. Current standard regimens are highly effective under controlled trial conditions with relapse rates of 5% or less. It is very unlikely better results can be achieved with new drugs, a clinically more relevant goal would be a regimen of comparable efficacy to standard treatment but of substantially shorter duration. In order for a new regimen to be licensed, it will be necessary to demonstrate that it is of comparable efficacy to the standard regimen. An important issue will be the choice of the margin of non-inferiority which needs to be justified both on statistical and clinical grounds; non-inferiority could be falsely concluded if a trial was not conducted appropriately, with substantial losses to follow-up or unsatisfactory laboratory procedures. It is particularly important therefore that such trials are conducted with a high degree of rigor. Analyses should be performed both by intention to treat, which is conservative and therefore biased towards no difference, and on a protocol correct population. Similar conclusions would be required from both analyses. Substantial developments have been made in tuberculosis bacteriology in recent years enhancing our ability to diagnose and differentiate strains of M. tuberculosis. Many of these techniques affect the design of trials but have yet to be evaluated in that setting. Non-inferiority pivotal trials require that, as far as practicable, the same techniques are used as were employed when the trials assessing the standard regimen were conducted.