The present peer-to-peer trust models are too coarse and lack ability of synthesizing the feedback. This paper presents a new fine-grained trust model for P2P networks based on grey relation. The peers are divided into different regions according to interest and speciality, and the direct trust is gained by colligating the evaluation for all attributes of one service. A memory factor is introduced to describe the characteristic that trust trail off along with time. The grey relation is used to measure the accurateness of feedback. Analysis and experiments show that compared with existing models, the proposed model evaluates peer’s trust value more accurately and it has better capability to resist the attack by the dynamic malice peers and peer’s dishonest feedback.