New vaccine candidates for tuberculosis are beginning to enter clinical trials. In this review we discuss issues surrounding the design of these candidates, and the way they were screened in animal models. First, screening vaccines for their ability to attenuate inevitably fatal tuberculosis in immunologically naïve mice might be leading to the selection of inappropriate candidates. We need to screen vaccines for their ability to stop the development of progressive disease, since this is what they must achieve in man. A solution to this problem is proposed. Secondly, we point out that some mouse models of tuberculosis in laboratories in developing countries, where exposure to environmental mycobacteria is large, mimic neglected aspects of human disease more closely than do low-dose infections in hyper-susceptible immunologically naïve mice in the USA or Europe. We need to think more about geographical differences in immunological experience, and these mouse models can help us. Thirdly, we conclude that in developing countries where BCG fails this is not because there is too little Th1 response, but rather because the Th1 response is rendered ineffective and immunopathological by other subversive mechanisms, including IL-4 responses and inappropriate regulatory T cell function. Therefore, we suggest that vaccines that will work in those countries might need to have immunoregulatory properties that can switch off pre-existing subversive mechanisms, and block their development in the future. The development of such vaccines, that might work where BCG does not, will require a greater understanding of the roles of the many types of regulatory T cell in tuberculosis.