Online reputation systems are playing increasingly important roles in influencing people’s online purchasing/downloading decisions. The manipulations against such systems which overly inflate or deflate reputation scores of online items are evolving rapidly. For example, for just $9.99, a video on YouTube could receive 30 “I like” ratings or 30 real user comments provided by “IncreaseYouTubeViews.com”. Without proper defense schemes, attacks against reputation systems can overly inflate or deflate item reputation scores, crash users’ confidence in online reputation systems, and lead to economic loss. Based on the anomaly detection results, the further evaluate users’ trust values in this section. In most trust models, users’ trust values are determined only by their good and bad behaviors. However, it is not sufficient. Two trust calculation scenarios. First, user A has conducted 5 good behaviors and 5 bad behaviors. Second, user B is a new coming user and has no behavior history. In several trust models , both of their trust values will be calculated as 0.5, although we are more confident in user A’s trust value. To differentiate these two cases, the concept of behavior uncertainty is introduced by the Dempster-Shafer theory, to represent the degree of the ignorance of behavior history. In this work, we adopt the behavior uncertainty by proposing a trust model based on the Dempster- Shafer theory. Based on the anomaly detector, for each given item, we could determine which ratings are suspicious. We then define a user’s behavior value on a single item as a binary value to indicate.