Economic operation is a major concern for microgrids. Conventionally, economic dispatch of distributed generations (DGs) are solved by centralized control with optimization algorithms or distributed control with consensus algorithm. To improve the reliability, scalability and economy of microgrids, a fully decentralized economic power sharing strategy is proposed in this paper. The proposed method is based on frequency/incremental cost droop (f/IC) characteristics and incremental cost (IC) functions of DGs. ICs of DGs reach equality with the convergence of system frequency. Power dispatch of each DG is automatically achieved based on its relevant incremental cost function. Therefore, by using this method, the incremental cost of each DG will reach equality autonomously and the total operating cost can be optimized without any communication or central controllers. Simulation platform of an autonomous AC MG with three DGs is built in Matlab/Simulink to verify the effectiveness of the proposed method.