Reputation systems are widely applied in the electronic marketplace to provide reference information for consumers and alleviate their transactional risk. Despite the convenience of reputation systems, they are vulnerable to malicious ratings injected by fraudulent raters. Although several rating fraud detection methods have been proposed in previous research, their performance has limitations in relatively complicated attack scenarios. In this study, we have used real‐world rating datasets from Expedia.com, TripAdvisor.com, and Amazon.com to explore features of collaborative rating fraud. Leveraging these features, this study proposes a two‐phase approach for fraudulent rater detection. It first examines the rating series of each entity and identifies the suspicious entities. For raters of all suspicious entities, a clustering technique is then adopted to discriminate fraudulent raters. We have performed a series of empirical evaluations to compare the effectiveness of the proposed method with three benchmark methods using another real‐world dataset, in terms of fraudulent rater detection accuracy or rating difference. Our results show the advantages of the proposed method in various rating fraud scenarios.