Abstract: Grid resource optimized allocation is a NP problem, which can be solved by genetic algorithm efficiently.An improved genetic algorithm is proposed in order to overcome the defect of slow convergence speed in this algorithm.Qualified initial population and excellent members protecting crossover strategies are brought in, they all can help to enhance the searching ability and convergence speed.This method plays a practical role by considering many parameters such as the communication cost between the resources and tasks, execution time of task and the utility of resources, and describing resources uniformly in the way of combining the characteristics of grid resources.