This paper describes an approach for decomposition and reduction of dynamical models of largo scale power systems. In this approach the system is decomposed into two subsystems, the first one is described by a linear model while the second is described byanon-linear model. This decomposition is based on a derived criteria for linearization by which we can know the two subsystemsa priori. Further, the linear subsystem model is reduced by aggregation and hence the order of the system is reduced. The results of the validity of this approach as applied to two different largo power systems are indicated.