9407029 TRANGENSTEIN Advances in computer hardware have greatly extended the predictive capability of reservoir simulators. Even with current gigaflop performance, however, the capabilities of the machines executing state-of-the-art algorithms are still far below the needs; in order to reduce the risk in planning projects, the petroleum industry needs much greater computational power in order to make improved predictions of reservoir performance. Further advances can come by introducing sophisticated algorithmic techniques, such as adaptive local grid refinement and parallel processing. In this postdoctoral research program, we propose to extend adaptive local grid refinement techniques, currently being developed at Duke University for micellar- polymer flooding, to compositional reservoir simulation. We also propose to investigate the parallelization of this algorithm on a network of IBM RS6000 workstations at Mobil Exploration and Production Technical Center.