I investigate the effect of competition on bidder behavior and procurement cost using highway auction data from Michigan. While a bidder’s distance to a project location is important in explaining participation and bid levels, there is no evidence of more aggressive bidding when competitors are located close to the project. This pattern is at odds with theoretical predictions based on first-price auctions with private costs but can be rationalized by a more general model that allows firms to have nonindependent private information and partially common costs. I show that such a model is identified from observable variation in bidder-specific cost shifters, and develop an estimation procedure that exploits variation in project locations. The findings point to significant common costs and to a high degree of correlation in private information. Model estimates are used to show that common costs and information correlation reduce the effect of competition on procurement costs by 28 percent relative to a benchmark that assumes independent private costs. Moreover, subsidies to weak bidders are estimated to cost 27 percent more than in the private costs benchmark.