BACKGROUND: Randomized trials assessing BCG vaccine protection against tuberculosis have widely varying results, for reasons that are not well understood. METHODS: We examined associations of trial setting and design with BCG efficacy against pulmonary and miliary or meningeal tuberculosis by conducting a systematic review, meta-analyses, and meta-regression. RESULTS: We identified 18 trials reporting pulmonary tuberculosis and 6 reporting miliary or meningeal tuberculosis. Univariable meta-regression indicated efficacy against pulmonary tuberculosis varied according to 3 characteristics. Protection appeared greatest in children stringently tuberculin tested, to try to exclude prior infection with Mycobacterium tuberculosis or sensitization to environmental mycobacteria (rate ratio [RR], 0.26; 95% confidence interval [CI], .18-.37), or infants (RR, 0.41; 95% CI, .29-.58). Protection was weaker in children not stringently tested (RR, 0.59; 95% CI, .35-1.01) and older individuals stringently or not stringently tested (RR, 0.88; 95% CI, .59-1.31 and RR, 0.81; 95% CI, .55-1.22, respectively). Protection was higher in trials further from the equator where environmental mycobacteria are less and with lower risk of diagnostic detection bias. These associations were attenuated in a multivariable model, but each had an independent effect. There was no evidence that efficacy was associated with BCG strain. Protection against meningeal and miliary tuberculosis was also high in infants (RR, 0.1; 95% CI, .01-.77) and children stringently tuberculin tested (RR, 0.08; 95% CI, .03-.25). CONCLUSIONS: Absence of prior M. tuberculosis infection or sensitization with environmental mycobacteria is associated with higher efficacy of BCG against pulmonary tuberculosis and possibly against miliary and meningeal tuberculosis. Evaluations of new tuberculosis vaccines should account for the possibility that prior infection may mask or block their effects.