In order to address the challenges of complex process and low precision in traditional device modeling, several artificial neural network technologies are used to investigate the scattering parameters(S-parameters) of gallium arsenide pseudomorphic high electron mobility transistor (GaAs PHEMT) at different temperatures. At first, the S-parameters are randomly divided into training set and test set, which are modeled by double hidden layer conjugate gradient backpropagation neural network (CG-BPNN) and extreme learning machine (ELM), respectively. Then, the fitting results and error curves of the two models in predicting the S-parameters are given. The experimental results show that the CG-BPNN has the general fitting result with large errors in some data, whilemost of the data predicted by ELM can achieve ideal fitting result. In addition, the mean square error of CG-BPNN and ELM are 0.013 508 and 0.002 254 9, respectively. Through the above experiments, it is proved that ELM has better modeling effect on the S-parameters of GaAs pHEMT at different temperatures. Therefore, the proposed modeling method can accurately and stably characterize the S-parameters of GaAs pHEMT at different temperatures.