The protective effect of BCG against tuberculosis (TB) estimated in randomized controlled trials and observational studies ranges from negative to close to a 100%. One of the many explanations offered for this is that different immunological mechanisms may be associated with protective effect against different forms and sites of disease. In this investigation, we recalculated vaccine protective effect separately for pulmonary disease and for meningeal/miliary disease in randomized controlled trials and case-control studies, tested for heterogeneity in site-specific estimates of protective effect and calculated a summary measure when appropriate. We found protective effect against pulmonary disease to be heterogeneous to a statistically significant degree, and thus we did not calculate a summary measure of protection. Protective effect against meningeal and miliary TB was higher than against pulmonary disease and, except for a single study with two cases only, appeared to be homogenous. Summary BCG protective effect against miliary or meningeal TB in randomized controlled trials was 86% (95% confidence interval [CI] 65, 95) and in case-control studies 75% (95% CI: 61, 84). The fact that protective effect appeared to be homogeneous against meningitis and miliary TB but not against pulmonary disease may result from the fact that patients with meningitis are on average younger and thus less likely to have been exposed to atypical bacteria; to a waning of the protective effect of BCG; or from the diversity of mechanisms of pathogenesis of pulmonary disease, which can originate from reinfection, reactivation or primary progression.