A newly developed frequency-domain optimization approach for solving the model matching problem, perfect or imperfect, is presented. A new necessary and sufficient condition for perfect model matching in linear multivariable systems is derived, which, if satisfied, results in an exact model matching control system. If the perfect matching condition fails, an optimal approximation design can be used so that the controlled system responds in a manner approximating to the chosen model. Compared with existing methods, the proposed optimization method is conceptually simpler.