We estimate the effects of sulfur dioxide (SO2) exposure on short- and long-run mortality in the US over the period 1972-1988. Using wind direction as an instrument for changes in daily SO2 levels, we estimate that a one-day, one part-per-billion increase in SO2 raises all-age mortality by 0.18 deaths per million over the next four weeks. This immediate mortality increase is concentrated among the elderly. We then calibrate a dynamic production model of health using our short-run estimates. The model suggests that the 1972-1988 reductions in SO2 can explain a meaningful share of the observed mortality decline over this time period. (joint with Julian Reif, who is also visiting TSE).