A distributed coalition formation algorithm is presented to network structure in multi-agent systems. For agents, the task with the highest performance/cost ratio is first selected, which result in the potential coalition. If the resources of the coalition are greater than the required resources of the task, some agents quit to opt for tasks with the second highest performance/cost ratio. By so doing, the possibility for coalition formation can be raised. The simulation results show that compared with other methods, this algorithm can highly improve the efficiency and utility of the systems.