In this paper, three tasks division strategies of load-balancing are given, which include a round-robin fashion, a domain partitioning fashion and a fashion of round-robin combining with domain partitioning, and a comparison is made. Then tasks division methods are proposed which can reduce solving times and benefit load-balancing and increases parallel efficiency for a simulation of reservoir of multiple layers two-dimension two-phase flow numerical problems on distributed memory parallel systems. This approach is applied to a large scale reservoir simulation which has over 720 000 grid points. The practical results show that the tasks generated by the strategy are well-balanced and benefit to improve speedup. This approach is apt to the problems for domain parallel or data parallel.