Online freelance marketplaces are websites that match buyers of electronically deliverable services with freelancers. Although freelancing has grown in recent years, it faces the classic "information asymmetry" problem.buyers face uncertainty over seller quality. Typically, these markets use reputation systems to alleviate this issue, but the effectiveness of these systems is open to debate. We present a dynamic structural framework to estimate the returns to seller reputations in freelance sites. In our model, a buyer decides in each period whether to choose a bid from her current set of bids, cancel the auction, or wait for more bids. In the process, she trades off sellers' price, reputation, and other attributes, as well as the costs of waiting and canceling. Our framework addresses dynamic selection, which can lead to underestimation of reputation, through two types of persistent unobserved heterogeneities: bid arrival rates and buyers' unobserved preference for bids. We apply our framework to data from a leading freelance firm. We find that buyers are forward looking, that they place significant weight on seller reputation, and that not controlling for dynamics and selection can bias reputation estimates. Using counterfactual simulations, we infer the dollar value of seller reputations and provide guidelines to managers of freelance firms.