An experimental approach is used to examine the performance of three different multi-unit auction designs: discriminatory, uniform-price with fixed supply, and uniform-price with endogenous supply. We find that the strategies of the individual bidders and the aggregate demand curves are inconsistent with theoretically identified equilibrium strategies. The discriminatory auction is found to be more susceptible to collusion than are the uniform-price auctions, and so contrary to theoretical predictions and previous experimental results, the discriminatory auction provides the lowest average revenue. Consistent with theoretical predictions, bidder demands are more elastic with reducible supply or discriminatory pricing than in the uniform-price auction with fixed supply. Despite a lack of a priori differences across bidders, the discriminatory auction results in significantly more symmetric allocations.