BCG protection varies and in some places (nearest the equator) is low or absent. Understanding this variation can inform the efforts to develop new vaccines against tuberculosis. Two main hypotheses are used to explain this variation: under masking, new vaccines are unlikely to increase protection; under blocking new vaccines have a greater potential to be effective when BCG is not. We conducted a cluster randomized trial to explored the masking and blocking hypotheses by studying BCG vaccine efficacy of neonatal vaccination and when administered for the first or a second (revaccination) time at school age in two sites (Manaus close and Salvador further south from the equator). Seven hundred and sixty three state schools were matched on socio economic characteristics of the neighborhood and 239,934 children were randomized to vaccine (BCG vaccination at school age) or control group. Protection by first BCG vaccination at school age was high in Salvador (34%, 95% CI 7-53%, p=0.017) but low in Manaus (8%, 95% CI t0 39-40%, p=0.686). For revaccination at school age, protection was modest in Salvador (19%, 95% CI 3-33%, p=0.022) and absent in Manaus (1%, 95% CI to 27-23%, p=0.932). Vaccine efficacy for neonatal vaccination was similar in Salvador (40%, 95% CI 22-54%, p<0.001) and Manaus (36%, 95% CI 11-53%, p=0.008). Variation in BCG efficacy was marked when vaccine was given at school age but absent at birth, which points towards blocking as the dominant mechanism. New tuberculosis vaccines that overcome or by pass this blocking effect could confer protection in situations where BCG is not protective.