The Clean Air Act requires the EPA to conduct annual auctions of emission allowances. Under the discriminative auction rules, sellers with the lowest asking prices receive the highest bids. This paper studies an inverted version of this auction in which buyers face the same incentives as sellers in the EPA auction. Consistent with theoretical predictions, buyers bid above their valuation, auction outcomes are inefficient, and increasing the number of buyers increases bids. Buyers facing human opponents compete more aggressively than the risk-neutral prediction, but bids do not differ systematically from this prediction when buyers face computerized Nash "robots."