Review sites have become increasingly important sources of information for consumers. Because these reviews affect sales, businesses have the incentive to game the system by leaving positive reviews for themselves, or negative reviews for their competitors. Such review fraud undermines the trustworthiness of consumer reviews, and constitutes a major risk factor for review sites. In this paper, we investigate review fraud on the popular consumer review site Yelp. We construct a novel data set to analyze this problem, combining restaurant reviews with Yelp's algorithmic indicator of fake reviews. Using this imperfect indicator as a proxy, we develop an empirical methodology to identify the points in the life-cycle of a business during which review fraud is most prevalent. We find that a restaurant's changing reputation affects its decision to engage in review fraud. Specifically, a restaurant is more likely to seek a positive fake review when its reputation is weak, i.e., when it has few reviews, or it has recently received bad reviews. Consistent with theory, we find that chains are less likely than independent restaurants to engage in review fraud. We then turn our attention to negative review fraud, and find that increased competition by similar restaurants is the driving force behind it.