In this paper, an enhanced adaptive frequency sampling algorithm for robust electromagnetic (EM) modeling and computer aided design (CAD) is proposed. The algorithm is based on rational interpolation, which leads to more accurate RF/microwave models as compared to other existing interpolants, e.g. spline. Starting with a minimal number of support points i.e. EM data and with lowest-order rational functions, the algorithm systematically produces a rational function model, which meets a user-specified accuracy. In each stage of the algorithm, additional poles and/or zeros are added, and new support points are adaptively selected based on a statistical criterion. The algorithm distinguishes critical regions of an EM component's behavior that need to be densely sampled and uses relatively fewer data in other regions. The advantages of the proposed algorithm are shown via RF/microwave examples.