BACKGROUND: Until recently, evidence that protection from the bacillus Calmette-Guérin (BCG) vaccination lasted beyond 10 years was limited. In the past few years, studies in Brazil and the USA (in Native Americans) have suggested that protection from BCG vaccination against tuberculosis (TB) in childhood can last for several decades. The UK's universal school-age BCG vaccination programme was stopped in 2005 and the programme of selective vaccination of high-risk (usually ethnic minority) infants was enhanced. OBJECTIVES: To assess the duration of protection of infant and school-age BCG vaccination against TB in the UK. METHODS: Two case-control studies of the duration of protection of BCG vaccination were conducted, the first on minority ethnic groups who were eligible for infant BCG vaccination 0-19 years earlier and the second on white subjects eligible for school-age BCG vaccination 10-29 years earlier. TB cases were selected from notifications to the UK national Enhanced Tuberculosis Surveillance system from 2003 to 2012. Population-based control subjects, frequency matched for age, were recruited. BCG vaccination status was established from BCG records, scar reading and BCG history. Information on potential confounders was collected using computer-assisted interviews. Vaccine effectiveness was estimated as a function of time since vaccination, using a case-cohort analysis based on Cox regression. RESULTS: In the infant BCG study, vaccination status was determined using vaccination records as recall was poor and concordance between records and scar reading was limited. A protective effect was seen up to 10 years following infant vaccination [< 5 years since vaccination: vaccine effectiveness (VE) 66%, 95% confidence interval (CI) 17% to 86%; 5-10 years since vaccination: VE 75%, 95% CI 43% to 89%], but there was weak evidence of an effect 10-15 years after vaccination (VE 36%, 95% CI negative to 77%; p = 0.396). The analyses of the protective effect of infant BCG vaccination were adjusted for confounders, including birth cohort and ethnicity. For school-aged BCG vaccination, VE was 51% (95% CI 21% to 69%) 10-15 years after vaccination and 57% (95% CI 33% to 72%) 15-20 years after vaccination, beyond which time protection appeared to wane. Ascertainment of vaccination status was based on self-reported history and scar reading. LIMITATIONS: The difficulty in examining vaccination sites in older women in the high-risk minority ethnic study population and the sparsity of vaccine record data in the later time periods precluded robust assessment of protection from infant BCG vaccination > 10 years after vaccination. CONCLUSIONS: Infant BCG vaccination in a population at high risk for TB was shown to provide protection for at least 10 years, whereas in the white population school-age vaccination was shown to provide protection for at least 20 years. This evidence may inform TB vaccination programmes (e.g. the timing of administration of improved TB vaccines, if they become available) and cost-effectiveness studies. Methods to deal with missing record data in the infant study could be explored, including the use of scar reading. FUNDING: The National Institute for Health Research Health Technology Assessment programme. During the conduct of the study, Jonathan Sterne, Ibrahim Abubakar and Laura C Rodrigues received other funding from NIHR; Ibrahim Abubakar and Laura C Rodrigues have also received funding from the Medical Research Council. Punam Mangtani received funding from the Biotechnology and Biological Sciences Research Council.