With the increase in the popularity of online marketplaces, the relevance of customer reviews to business success has also increased. This is especially important in decentralized markets with no central controlling authority. Current online review systems are plagued by several problems, including lack of incentives for reviews, subjective reviews, and collusion between buyers and sellers to unduly influence seller reputation. In this paper, we propose methods to detect sybil attacks on marketplace reputation systems. Since sybil attacks are very complex to detect, especially in decentralized systems, we suggest a collection of metrics to detect the sybils, provide a new weighted system for potential sybils, and illustrate their efficacy using example scenarios. For three chosen threat models, we generate synthetic data. We also employ 26 machine learning classifiers that are trained and tested on the synthetic data. The initial results are quite encouraging. We conclude that a combination of techniques are necessary to detect complex sybil attacks on marketplaces.