A neural network approach to a system identification problem is presented. Traditional system identification techniques require knowledge of the model structure before parameter estimation methods can be applied. This approach requires less a priori information and the knowledge of model structure is not essential. A simulation study on a power system has demonstrated the application of this technique.