This study uses quasi-experimental variation from a public health program implemented in Romania that targeted Roma, Europe's largest and most disadvantaged ethnic minority. The program employed health mediators to increase the provision of information about already existing, free of charge health services available for children and pregnant women. We find that, in rural areas, the program led to large increases in prenatal care take-up rates but no improvements in children's health at birth. However, we find significant reductions in infant mortality caused by perinatal complications.