In a laboratory experiment, we compare two auction mechanisms that determine the sequence of service to queued customers. In the server-initiated auction, the server, when idle, sells the right to be served next to the highest bidding customer in the queue and distributes the proceeds among the remaining customers. We show that this mechanism has an efficient equilibrium. In the customer-initiated auction, new arrivals can sequentially trade places with queued customers. This mechanism does not have an efficient equilibrium. We use two novel experimental protocols to examine the behavioral properties of both auction mechanisms. We find that, on average, the server-initiated auction and the customer-initiated auction perform equally well in terms of efficiency gain. Moreover, participants indicate that they find the server-initiated auction a fairer mechanism than the customer-initiated auction. When voting between the two auctions, participants tended to favor the server-initiated auction. We also find evidence of endowment and sunk-cost effects, which partially explains deviations from standard theory predictions.