The time consumed during the execution of queries in a parallel and distributed environment is highly affected by the form in which the tables comprising a database have been fragmented. The classical methods of fragmentation in a distributed database system helps, to a great extent, to make the information retrieval faster and with smaller calculation efforts. This is particularly true for applications where the specifications are well known in advance at the time of the creation of the tables that compose the database, which in some form or other influenced the design and the definition of the type of fragmentation and their distribution on different sites of processing. Nevertheless, the above characteristics cannot be used in applications where the distributed management cannot do inferences that help it to know in what sites the data with the characteristics the user is looking for are located. Under these conditions the time and amount of work used by the participants in the query solution can be highly increased. In this paper the dynamic virtual fragmentation method is presented. The method works as an alternative way that allows us to diminish response time consumed by queries using horizontal fragmented tables. The proposed method has been proven and implemented in a database parallel server running on a shared nothing supercomputer.