A reliable methodology for accurate modeling of microwave filter is presented. Our approach exploits co-kriging that utilizes low-fidelity and high-fidelity electromagnetic simulation data and combines them into a single surrogate model. Densely sampled low-fidelity data determine a trend function, which is further corrected by sparsely sampled high-fidelity simulations. Low-fidelity electromagnetic data are also enhanced by using a frequency scaling to reduce its misalignment with the high-fidelity model. With our method, accurate models can be obtained at a fraction of the cost required by conventional approximation models that are exclusively based on high-fidelity simulations. Three examples of microstrip filters are considered for verification purposes. We also provide comparisons with conventional approximation models and include an application of co-kriging models for filter design optimization. Copyright © 2013 John Wiley & Sons, Ltd.