In this paper, I present evidence that current economics research significantly underestimates the effects of air pollution, regardless of the outcome of interest. This bias exists even in quasi-experimental estimates and arises from the way researchers define individual-level pollution exposure. A polluter’s effect on nearby residents changes dramatically with the direction of the wind, and most popular methods, including geographic diff-in-diffs and monitor-based interpolations, are unable to account for such sharp changes in exposure over short distances. To solve this problem, I use an atmospheric dispersion model, which explicitly accounts for meteorological conditions, to determine the effect of every polluting firm on every house in greater Los Angeles. I then estimate the effect of NOx emissions on house prices and neighborhood composition using the exogenous variation in emissions caused by the California Electricity Crisis of 2000 and a cap-and-trade program in greater Los Angeles. The estimated price response is much larger than past estimates and implies that the social value of the cap-and-trade program is roughly \$502 million per year, 15 times larger than the associated abatement costs. However, when based on conventional measures of pollution exposure, this estimated valuation is small and statistically indistinguishable from zero. The estimated neighborhood sorting response suggests that, despite the high aggregate value, low-income households may not have benefited much from the improvement in air quality.