Air pollution poses serious risks to public health. Combatting it sensibly requires credible empirical estimates of the relevant costs and benefits. The intent of this thesis is to provide such estimates. Chapter 1 examines the value of reducing emissions from power plants, an important source of air pollution in several countries. Within a model of health, consumption, production, power generation, and resource extraction, I derive a 'sufficient statistics' formula for the change in social welfare due to a small reduction in emissions. The formula simplifies to a comparison of marginal benefits (in terms of reduced mortality risk, monetized using the value of a statistical life) and the marginal cost of abatement. I estimate these inputs using quasi-experimental variation induced by the Clean Air Interstate Rule, a policy that capped power plant emissions in the United States. Results indicate that further reducing those emissions would be worthwhile. Chapter 2 re-examines the effect of a county's regulatory status under the US Clean Air Act on the change in its air pollution concentration, the 'first-stage' underlying causal estimates of the benefits of reducing air pollution in several studies. Using data thought no longer to exist, I find that one of the commonly-used measurement approaches -- a regression-discontinuity estimator -- is invalid. The other commonly-used approach -- a difference-in-differences estimator -- delivers inflated estimates of the effects of regulation on air pollution. These findings suggest that the literature substantially understates the benefits of reducing air pollution. Chapter 3, joint with Robert McMillan, provides causal estimates of the effects of sustained exposure to severe air pollution on mortality risk. Our research design is based on the 'smoke control' provisions of the UK's Clean Air Act of 1956, which granted local authorities power to address emissions from the domestic chimney. We find that smoke control caused significant reductions in air pollution in areas that implemented it relative to those that did not. Our 2SLS estimates, which combine this exogenous variation in air quality improvements with local mortality data, indicate a significant reduction in probability of death. They are relevant when calculating air pollution's costs in coal-burning middle-income countries today.