We compare a sealed-bid uniform-price auction (the Treasury’s experimental format) with a sealed-bid discriminatory auction (the Treasury’s format heretofore), assuming the good is perfectly divisible. We show that the auction theory that prompted the experiment, which assumes single-unit demands, does not adequately describe the bidding game for Treasury securities. Collusive strategies are self-enforcing in uniform-price divisible-good auctions. In these equilibria, the seller’s expected revenue is lower than in equilibria of discriminatory auctions.