The core idea of the credibility evaluation method of electromagnetic simulation results is to replace the experts with electromagnetic computing professional background to evaluate the credibility of simulation results. The representative algorithm is the FSV method (Feature Selective validation, FSV) proposed by the IEEE Standards Association. However, the existing credi-bility assessment methods all use statistical indicators or signal processing methods to simulate the real thoughts of experts, and have not achieved true artificial intelligence. In this paper, a cred-ibility evaluation method of simulation results based on convolu-tional neural network is proposed, which aims to integrate the real ideas of experts (background knowledge of electromagnetic calcu-lation) into the evaluation, instead of just mechanical numerical calculation, and to avoid evaluation errors caused by nonprofes-sional.