A new adaptive multi-dimensional RF/microwave modeling algorithm based on rational interpolants is proposed. For the first time, the algorithm integrates rational functions, genetic concepts, and frequency mapping into a unified CAD tool, which facilitates accurate and fast EM based RF/microwave modeling. Given an RF/microwave modeling problem, highly non-linear and relatively less nonlinear model inputs are distinguishable. The algorithm consists of two phases. Starting with a minimal number of EM data and with lowest-order rational functions, Phase 1 systematically produces a sub-model with highly non-linear inputs alone. In Phase 2, an initial overall model is first constructed based on the above sub-model. A hybrid mapping then fine-tunes the initial model to capture the RF/microwave tendencies w.r.t. the less non-linear inputs. The algorithm is illustrated through a practical example.