For deception jamming countermeasures of multistatic radar systems, existing intelligent anti-jamming methods require sufficient training samples and a large amount of labelled data, but it is difficult to obtain abundant labelled radar echo data in realistic operational environments. A deception jamming discrimination method based on semi-supervised learning with generative adversarial networks is proposed to specifically handle the situation of inadequate labelled samples. In this way, a small amount of labelled data and a large amount of unlabelled data obtained from radar stations, together with pseudo-labelled data generated by the generator are used to train the discriminator to improve the performance of jamming discrimination and the robustness of the discrimination network by exploiting the game between the generator and the discriminator. Simulation results show that, the proposed method can achieve the same performance using less than 10% of the labelled data of existing algorithms. It reduces data requirements and enhances operational capabilities, which is better suited to real-world battlefield environments.