A distributed system database performance is strongly related to the fragment allocation in the nodes of the network. An heuristic algorithm for redistributing the fragments is proposed. The algorithm uses the statistical information relative to the requests send to a distributed database. This algorithm minimizes the size of the data transferred for solving a request. Assuming that a distribution of the fragments in the nodes of a network is known, the algorithm generates a plan to transfer data fragments, plan that will be used to evaluate a request.