Abstract-Presently, thermopile infrared sensors have been successfully commercialized in different application areas such as contactless temperature measurements and infrared detectors. In the process of designing the application circuit of thermopiles, circuit simulation is necessary to improve the efficiency of the circuit design. This work reports a method based on neural networks to model a thermopile infrared sensor, by which the fundamental characteristics including the temperature-voltage output characteristic, the frequency response characteristic, the angle characteristic and the optical characteristic are modelled in PSpice. Firstly, three appropriate neural networks are created and trained to approach each input-output characteristic respectively, after which the structures, weights and biases of each neural network are acquired. Secondly, the obtained structures are described with the PSpice programming language to establish sub-circuits respectively. Finally, these sub-circuits are integrated into a unitary circuit based on the relationship between the inputs and output of the sensor to form the final model.