With the exponential growth of data scales in the contemporary e-commerce systems, rating items with biased or misleading scores lead to poor performance of recommendation systems. Measures on user reputation are highly preferred to identify those with deliberate biased or random rating spammers. Despite the fact that previous methods are relatively feasible, they are not accurate or robust when the numbers of malicious users have reached a critical value. In this paper, we propose an iterative deviation-based user reputation ranking (IDR) method. It is inspired by the common fact that user with higher ranking usually performs less biased rating scores. Another factor that influences the ranking coming from their rating patterns. High quality rating scores are usually given by users with peaked rating patterns. Experimental results on four real sparse data sets show that the accuracy and robustness of the proposed method are better than the existing state of arts methods.