complete, efficient and stable method to ensure the safety of the network transactions. Trust management is one of the approaches having been developed these years. To improve the recommendation accuracy and reflect its positive or negative impact on network transactions, a new trust management model for P2P system is proposed in this paper. This model makes peer's recommendation more reasonable and precise by introducing a recommendation confidence interval, which is used to calculate the allowed range of the unsuccessful recommendation number. The model recounts the recommendation value which is provided by the other peers before the transaction occurs to meet the peer's practical requirements. A punishment mechanism is enabled when the unsuccessful recommendation number is accumulated to a given value. The experimental results show the model can more effectively use other peers' recommendation, reasonably take into account all kinds of transaction data, improve recommendation accuracy and facilitate the transactions safely.