Abstract: In order to build the electromagnetic pulse response models for surge protectors, a time-domain model based on the NARX neural network is given. The theoretical basis and design steps of modeling with the NARX neural network are presented. The transmission-line pulse system and electrostatic discharge system are built for proving the method. The measurement results obtained in transient voltage suppressor experiment is taken to set up the NARX neural network model. When the input signal is given as a square pulse, machine-model or body-metal-model electrostatic discharge electromagnetic pulse, the established model can forecast performance parameters such as response time, pulse peak and clamping voltage, which verifies the validity of the model.