Background: it is widely assumed that individuals with Mycobacterium tuberculosis (Mtb) infection remain at lifelong risk of tuberculosis (TB) disease. However, there is substantial evidence that self-clearance of Mtb infection can occur. We infer a curve of self-clearance by time since infection and explore its implications for TB epidemiology. Methods and findings: data for self-clearance were inferred using post-mortem and tuberculin-skin-test reversion studies. A cohort model allowing for self-clearance was fitted in a Bayesian framework before estimating the lifetime risk of TB disease and the population infected with Mtb in India, China and Japan in 2019. We estimated that 24.4% (17.8-32.6%, 95% uncertainty interval (UI)) of individuals self-clear within 10 years of infection, and 73.1% (64.6-81.7%) over a lifetime. The lifetime risk of TB disease was 17.0% (10.9-22.5%), compared to 12.6% (10.1-15.0%) assuming lifelong infection. The population at risk of TB disease in India, China and Japan was 35-80% (95% UI) smaller in the self-clearance scenario. Conclusions: the population with a viable Mtb infection may be markedly smaller than generally assumed, with such individuals at greater risk of TB disease. The ability to identify these individuals could dramatically improve the targeting of preventive programmes and inform TB vaccine development, bringing TB elimination within reach of feasibility.