The paper estimates the effect of SO2 pollution on infant mortality in Germany, 1985-2003. To avoid simultaneity problems, I exploit the natural experiment created by the mandated desulfurization at power plants, with wind directions dividing counties into treatment and control groups. Instrumental variable estimates are larger than conventional estimates and translate into an elasticity of 0.08-0.13. The observed reduction in pollution implies an annual gain of 895-1528 infant lives. Estimates are robust to controls for economic activity, climate, reunification effects, rural/urban trends and TSP pollution and are comparable across subsamples. Excess mortality mainly accrues in the first month after birth.