Replication is an effective method that improves data access in Data Grids; however, the improvement of the efficiency of repilication is a crucial problem. Previous works on the replication mechanism mostly select high value replicas to replicate, and simply choose the files that have been accessed based on the access records of files. This paper starts with the analysis of the file access characteristics in the virtual organization (VO). After introducing the concept of implicit high-value file (IHVF), it proposes the cooperative replica prefetching mechanism (CoRPM) for virtual organization, which is based on which local grid nodes can obtain the replicas of IHVFs through cooperation with other gird nodes in the same VO. The architecture of CoRPM is presented first, in which prefetching elements running on every gird node work cooperatively to provide a file prefetching service for all the grid nodes in the VO. Then, on the basis of the design of CoRPM, the process of CoRPM is described, whose core algorithms include a job-type centric file prefetching algorithm and a prefetching file selecting algorithm. In the end, the paper evaluates the performance of CoRPM through simulations, and the results show that CoRPM does reduce the file access latency more effectively.