There is an urgent need for an improved TB vaccine. Vaccine development is hindered by the lack of immune correlates and uncertain predictive value of preclinical animal models. As data become available from human efficacy trials, there is an opportunity to evaluate the predictive value of the criteria used to select candidate vaccines. Here we review the efficacy in animal models of the MVA85A candidate vaccine in light of recent human efficacy data and propose refinements to the preclinical models with the aim of increasing their predictive value for human efficacy.