To address an electric vehicle’s magnetic emission problem, a model-based improvement strategy is proposed to avoid resource-intensive experimental diagnosis processes, thus achieving higher efficiency. Considering the electrical and structural characteristics of electric vehicles, a network model is developed to predict magnetic emissions. It decomposes the electronic power system into a global network and external circuit nodes according to electrical size. The Z-parameter is used to characterize the global network for the decomposition of impedance coupling so that the model parameters can be obtained separately using different methods. With this network model, an evaluation index is designed to measure the influence of technical factors on magnetic emissions by comprehensively considering their contributions and rooms for improvement. Engineers can directly determine the main interference source according to this evaluation score, and select a proper filter to attenuate the interference.