Non-adherence to tuberculosis (TB) treatment is a major challenge to global TB control because it increases the risk of treatment failure, relapse and the emergence of drug-resistant TB. Although the problem is widely acknowledged, there is still no clarity about the exact impact of different levels and patterns of non-adherence on treatment outcome. This hampers the provision of adequate advice to patients and clinicians, and it challenges the development and evaluation of adherence-promoting interventions. In this article, we explain why we are still in the dark with respect to predicting how different types of non-adherence to TB treatment affect treatment outcome. We show that we lack uniformity in how we define and measure non-adherence, that we have no easily accessible treatment success indicators, and that the relationship between treatment adherence and outcome is influenced by a number of pathogenic, immunological and pharmacological factors that are only partly understood. We conclude that an integral 'bench and bedside approach' that incorporates experimental studies with in vitro models and animals, as well as observational studies in patients with TB, is needed to help us get out of the dark regarding the adherence-response relationship in TB treatment