The authors investigate the problem of object allocation in a distributed environment with virtually replicated data. The traditional approach to improving data availability in a distributed system is to replicate data. A high degree of replication, however, imposes a serious burden to the system when updates are performed. Data inference can be used to reduce the degree of replication in the system while still providing high data availability. A model to allocate objects under such an environment is proposed. Rules based on application semantics are developed to reduce the search space for optimal allocation. Heuristic algorithms are proposed for allocation when the reduced search space is still prohibitively large. Examples are given to illustrate the effectiveness of the proposed algorithms.>