We analyze a new auction format in which bidders pay a fee each time they increase the auction price. Bidding fees are the primary source of revenue for the seller, but produce the same expected revenue as standard auctions. Our model predicts a particular distribution of ending prices, which we test against observed auction data. Our model fits the data well for over three-fourths of routinely auctioned items. The notable exceptions are video game paraphernalia, which show more aggressive bidding and higher expected revenue. By incorporating mild risk-loving preferences in the model, we explain nearly all of the auctions.