In recent years, research attention on smart grid comprising distributed power generators has increased. To produce electricity in the smart grid, many micro grids (MGs) may exploit various renewable energy resources. Because the production capacity of renewable resources cannot be controlled, the MGs often require the power plants to provide power for them. However, thepower loss between each MG and the power plant is larger than that among the MGs. To alleviate this power loss, we propose a game theoretic coalition formulation strategy for the MGs dubbed GT–CFS. Our proposed GT–CFS allows the MGs (belonging to the same macro station (MS)) to autonomously cooperate and self-organize into a partition composed of disjoint MGcoalitions. Also, GT–CFS enables the MGs, in a distributed manner, to decide whether they will remain in the coalitions or not upon environmental changes, e.g., the variation of the powerdemand of the MGs. Within every coalition, MGs coordinate the power transfer among themselves as well as with the MS, in a fashion to optimize a utility function, which captures the total losses over the power distribution lines. MGs in the same coalition will distribute the extra profits (i.e., payoff) produced from forming coalitions by their “Shapley value.” Through computer simulations, we demonstrate that the proposed GT–CFS reduces the average power loss per MG significantly in contrast with the conventional noncooperative approach.