This paper shows how structured model order reduction can be applied to power systems. For power systems divided into a study area and an external area, the proposed algorithm can be used to reduce the external area to a low order linear system, while retaining the nonlinear description of the study system. The reduction of the external area is done in such a way that the study system is affected as little as possible. It is shown that a lower model order can be attained when information about the study system is taken into consideration, than if the external system is reduced independently of it