Abstract: The reliability of data-nodes and the reliability of relevant network links were analyzed, and then the reliability model of replica service of cloud storage systems was constructed. According to the relationships among access reliability, the number of replicas and the number of user’s accesses, the reliability of data service and the trigger mechanism of replica generation were presented, and the storage node selection was aptly checked, then the replica distribution algorithm and replica deletion algorithm were proposed. Finally a series of experiments were conducted in the cloud storage system, named ERS-Cloud, and the results indicate that the approach can ensure the reliability of data service, and further decrease the number of replicas of the redundant storage.