Multi-input multi-output (MIMO) model reduction is essential for applying many modern control design methods in power systems. The challenges of MIMO model order reduction of the modern power grid increases with the inclusion of renewables such as inverter-interfaced wind farms, which introduces complexity in such power system models. The difficulty in reducing such models using gramian-based and modal truncation approaches is demonstrated, whereas a relatively new Interpolatory approach of the Iterative Rational Krylov Algorithm (IRKA) shows promising results. Next, a heuristic-based IRKA is proposed to improve the accuracy of model order reduction of modern grids with explicit preservation of the “critical modes” of the system. Tests on a 16-machine New England–New York system with two doubly-fed induction generator-based wind farms and a larger Brazilian system model demonstrate the effectiveness of the proposed approach.