In this paper, a coherency-based dynamic reduction method for power system stability studies is proposed. The proposed method is based on the principle that both short circuit current at each retained node and operating condition at boundary nodes after reduction is matched to those of original network. Two types of any radial and loop configured external network can be reduced, and each line constant and nodal power condition for those reduced networks can be automatically determined from the principle point of view stated above.The one of notable features of the proposed method is that generator internal voltage angle after the reduction can be specified as mean value of those angles in original network for radial reduction, and another one is that no negative branch impedance solution can be given for looped reduction while negative impedance is often resulted in conventional method.The proposed method is applied to both the IEEJ West-30 standard model and realistic bulk power system model, and its sufficient accuracy is verified by comparing transient stability swing to conventional method.