Aiming at the problems that the traditional judgment methods of the main offensive direction of air attack have higher accuracy and completeness requirements on the evaluation index system, and their adaptability to some complex conditions, such as fuzzy information, missing and contradictory information is not strong, the convolution neural network is introduced into the main offensive direction judgment model. With its strong ability of nonlinear modeling and its strong adaptability to fuzzy and noisy information, the implied intuitive thinking of the person, such as experience, knowledge, and so on, is obtained through the study of large sample set. The experiments show that the new evaluation model has strong fault tolerance and good robustness.