In auctions of complex and valuable assets and in some procurement auctions, a two stage procedure is commonly used. In the first stage, bidders make non-binding "indicative bids" that are used to separate serious candidates. Then, after an information gathering stage, serious bids are made. This paper develops a theory of indicative bidding based on the assumption that learning valuations is costly. In various models, we show that efficient entry is unlikely to be achieved under the current practice of indicative bidding. We analyze alternative rules that may improve upon the current design. In particular, we show that by introducing auctions for entry rights with binding first-round bids, the auctioneer may in general induce efficient entry and improve the performance of two-stage bidding. Conditions under which these alternative schemes work are identified, and the optimal auctions are characterized. We also show that a revenue equivalence result holds in this two-stage auction environment.