This dissertation empirically investigates the performance of different selling mechanisms. The first chapter addresses the tradeoff between auctions ans posted prices in the context of a scarce perishable good. It makes use of data on National Football League tickets offered at eBay and leverages the within-seller and within-product variation in mechanism choice to estimate a dynamic structural model in which heterogeneous, forward-looking sellers optimally choose between the different mechanisms and their features. Counterfactual results suggest that sellers would experience an average 87.75% decrease in expected revenues if auctions were removed and an almost 13% decrease if posted prices were. In turn, consumers would benefit from an auction-only platform since the expected number of transactions would increase and expected transaction prices would decrease. The second chapter is co-authored with Dom Coey, Brad Larsen, and Kane Sweeney and it focuses on online auctions. In particular, the majority of these auctions are effectively second-price auctions, which implies that to implement the optimal mechanism the seller's problem boils down to choosing the optimal reserve price. Hence, we propose a set of estimators to discover what the optimal reserve price should be and which remain valid under general conditions. We establish the asymptotic properties of these estimators, propose a learning policy for sellers who run absolute auctions to gather data to estimate what reserve price they should use, and illustrate our tools using data from eBay on smartphones. Finally, we also discuss how the approach we propose can be extended to explicitly account for asymmetric bidders and implement the Myerson (1981) optimal auction. The final chapter is also co-authored with Dom Coey, Brad Larsen, and Kane Sweeney. It addresses the empirical analysis of ascending auctions, which is often complicated by severe data restrictions. In particular, we extend the methodology proposed by Aradillas-López et al. (2013) by first proving it remains valid under bidder asymmetries even when the identities of bidders are not observed and then by showing how this information, when available, can be used to improve upon existing tools. We demonstrate the advantages of this approach using USFS timber auctions.