In this paper we present a new method to efficiently optimize small-signal equivalent circuits for microwave and millimeter-wave FET linear circuit design. The method couples the stochastic search of a Partially Elitistic Genetic Algorithm with a local search procedure. Up to 19 equivalent circuit elements have been included in the small-signal model for completeness and flexibility. Optimization examples are given for an ion-implanted MESFET up to 12 GHz, a pseudomorphic HEMT up to 50 GHz, and for synthetic data. The results show that the proposed algorithm is able to consistently provide an excellent fit between measured and calculated S-parameters without any need of a careful initial guess for the circuit element values. Also, once the device parasitics have been de-embedded, the algorithm is able to extract unique, physically meaningful values for the intrinsic device parameters, and it is numerically shown not to be affected by measurement uncertainties.