Current reputation schemes, employed by an existing P2P file sharing network are faced with many threats, such as aggregate feedback, conspiracy, and fake transactions, which have affected the performance of whole system. To protect the P2P file sharing network, this paper proposes an accurate and adaptive reputation mechanism (AARep). In the paper, after an elegant analysis, it is concluded that besides the transaction evaluation, the relative and underlying transaction information plays an important role in the reputation system. This paper makes two important contributions. First, trust value computation is enhanced by adopting a decay function to display the importance of every transaction in sequence order. This weeds out suspected nodes that are less similar and utilizes the confident factors of reflect reliability in all observed values. Second, to make the confidence factors work, a simple transaction validation protocol is developed. Experimental results show that AARep can significantly eliminate or minimize the effects of a verity of attacks and improve the overall performance effectively.