Several researchers have looked into various issues related to automatic parallelization of sequential programs for multicomputers. But there is a need for a coherent framework which encompasses all these issues. In this paper we present a such a framework which takes best advantage of the multicomputer architecture. We resort to tiling transformation for iteration space partitioning and propose a scheme of automatic data partitioning and dynamic data distribution. We have tried a simple implementation of our scheme on a transputer based multicomputer  and the results are encouraging.