The methodology used to calculate dynamic equivalents has three basic steps: the identification of the coherent groups of generating units, the network reduction and the dynamic aggregation of coherent generator models. Among these steps, the last one has not been studied lately and deserves more development. This disseration deals with the problem of dynamic aggregation of excitaion system models to calculate coherency-based dynamic equivalents for power system transient stability studies. The main objective is the determination of a criteria to choose the best equivalent model for a given composition of excitation system models in a group of coherent generators. The parameters of the excitation system equivalent model for cach group of coherent generating units are numerically adjusted to match the frequency response of the individual models. This is a multivariable otmization problem, which is solved using the Levenberg-Marquardt method. The dynamic performance of the proposed equivalents is evaluated in a test system.