Over the past 50 years, the Mycobacterium bovis bacille Calmette–Guérin (BCG) vaccine against tuberculosis (TB) has maintained its position as the world's most widely used vaccine, despite showing highly variable efficacy (0–80%) in different trials. The efficacy of BCG in adults is particularly poor in tropical and subtropical regions. Studies in animal models of TB, supported by data from clinical BCG trials in humans, indicate that this failure is related to pre-existing immune responses to antigens that are common to environmental mycobacteria and Mycobacterium tuberculosis. Here, we discuss the potential mechanisms behind the variation of BCG efficacy and their implications for an improved TB vaccination strategy.