This paper introduces a new procedure, based on linearized large-signal vector measurements, for extracting a nonlinear behavioral model for two-port active microwave devices. The technique is applied to a model structure that assumes a short-term memory condition and is formulated as a parallel connection of a limited number of frequency-weighted static nonlinearities. The proposed method consists of integrating the time-varying linear characterization of the device driven into a nonlinear state by a large signal. The experiment design and measurement setup are based on a large-signal network analyzer and are discussed in detail. In the second portion of this paper, insight is provided on the most meaningful model parameters, along with an extensive independent experimental validation, which considers a GaAs pHEMT as a case study and includes two-tone large-signal data, a wideband code division multiple access signal, bias-dependent -parameters, and dc data.