Abstract-To model nonlinear device behavior at microwave frequencies, accurate large-signal models are required. However, the standard procedure to estimate model parameters is often cumbersome, as it involves several measurement systems (dc, vector network analyzer, etc.). Therefore, we propose a new nonlinear modeling technique, which reduces the complexity of the model generation tremendously and only requires full two-port vectorial large-signal measurements. This paper reports on the results obtained with this new modeling technique applied to both empirical and artificial-neural-network device models. Experimental results are given for high electron-mobility transistors and MOSFETs. We also show that realistic signal excitations can easily be included in the optimization process.