Computer simulation is effective for the optimization of complex system. However, it is time-consuming. Therefore, the number of simulations must be strictly confined. To generate a good simulation plan, an optimized Latin hypercube sampling method is put forward. The method can not only reduce the number of simulations, but also ensure the generated plan of good orthogonality and proportional spacing. Cholesky decomposition was borrowed to generate initial solution, and simulated annealing algorithm was used to get optimized array. Also a dynamic weight parameter was defined to balance different optimization objectives. Finally, an example was constructed. The result shows that the optimized Latin hypercube sampling method can generate a sound simulation plan with less simulation.