This paper proposes a framework for demand estimation with data on bids, bidders' identities, and auction covariates from a sequence of eBay auctions. First the aspect of bidding in a marketplace environment is developed. Form the simple dynamic auction model with IPV and private bidding costs it follows that if participation is optimal the bidder searches with a reservation bid for low-price auctions. Extending results from the empirical auction literature and employing a similar two-stage procedure as has recently been used when estimating dynamic games it is shown that bidding costs are non-parametrically identified. The procedure is tried on a new data set. The median cost is estimated at less than 2% of transaction prices.