Abstract: Resource allocation in grid is an NP-hard problem. To optimize the grid system, a performance QoS optimization model is developed for grid task scheduling and objective function, with the number of resources and tasks analyzed in detail. Then, an improved ant colony algorithm named RACO(reputation-based ant colony algorithm) is presented to schedule tasks in grid, based on the dynamic reputation. Introducing a dynamic scheduling factor involving both space and time efficiencies, a local and global pheromone updating strategy is applied to RACO. Simulation results showed that RACO algorithm outperforms the conventional Min-min, Max-min and ACO in resource utilization rate and dynamic equilibrium.