A generalized shape-preserving response prediction (GSPRP) methodology for low-cost and accurate modeling of microwave structures is presented. Our approach exploits kriging interpolation to model a set of characteristic point of the response of interest (e.g., S-parameters versus frequency), which are subsequently used to restore the entire response. As the locations of characteristic points are easy to model, GSPRP ensures good accuracy without using large amounts of training data, which is in contrast to conventional approximation techniques. The proposed technique is demonstrated through a comprehensive numerical study as well as comparison with other modeling techniques. Application examples are also given.Index Terms-Computer-aided design (CAD), surrogate modeling, microwave component modeling.