Abstract: In the field of aircraft design, various methods for obtaining aerodynamic data have their own advantages and disadvantages, making it a challenge to accurately predict an aircraft's aerodynamic characteristics using a single approach. Therefore, in practical engineering applications, it is often necessary to fuse data from multiple sources to achieve a more accurate and comprehensive description of aerodynamic characteristics. In response to this need, this study takes the typical jet aircraft as an example and employs real flight data, simulated flight data, and Computational Fluid Dynamics (CFD) data. By combining these with deep neural networks, we propose a dual-level deep evidential fusion algorithm for aerodynamic data under epistemic uncertainty. In this algorithm, two standard confidence distribution methods are introduced, and the output of deep neural networks is combined with variational Dirichlet distribution parameters to express and quantify epistemic uncertainty during the model fusion process. Utilizing the Dempster-Shafer theory, this algorithm effectively fuses data from different sources and their associated uncertainties. The results indicate that this algorithm successfully fuses multi-source aerodynamic data, producing outcomes that not only conform to physical laws but also provide more accurate and comprehensive aerodynamic data, which demonstrate significant advantages over single-source data methods.