This dissertation develops a method to detect collusion and estimate its effect on the seller's revenue in first-price auctions with independent, private valuations. The challenge is that collusion may be difficult to detect because colluders can use a simple and costless strategy to make their bids appear competitive. If the econometrician observes an exogenous shifter of the level of competition in the auction in addition to the winning bids, a statistical test for collusion by a given bidder can be formulated as a test of independence between the exogenous shifter and the valuations that rationalize its bids under the null hypothesis that it is not colluding. Simulations confirm this test performs well even when colluders attempt to disguise their behavior. I then adopt a multiple hypothesis testing framework to simultaneously test for collusion bidder by bidder. By controlling the probability of making one or more type I errors, the set of rejected hypotheses serves as a lower confidence bound on the set of colluders. To produce a lower confidence bound on the cost of collusion, I use consistent estimates of the bidders' valuation distributions to numerically solve for the seller's expected revenues in auctions with and without collusion. To provide an example of this identification strategy, I use exogenous variation in the reserve prices at British Columbia's timber auctions to estimate the extent of collusion in the years preceding a lumber trade dispute between the United States and Canada.