Objective To solve the shortage problem of flexography special color matching model. Methods On the basis of the nonlinear and self-learning characteristics in the BP Neural Network, this paper brought in the LMBP algorithm to improve the traditional BP algorithm and build the flexography special color matching model. At the same time, combining with the printing betas, we trained the model with Matlab software. Results Based on the analysis of training results, we concluded that although the BP algorithm with 17 notes in hidden layer could meet the expected requirements, the LMBP algorithm with 8 notes in hidden layer had higher precision and better approximation effect. Conclusion This model met the accuracy requirement and can be used in practice.