Mixed kernel fisher face feature extraction algorithm based on Maximum Margin Criterion is proposed, aiming at the existing problem of KFDA algorithm. First of all, the original data is projected to high-dimensional data space by nonlinear mapping. Then , the introduction of Maximum Margin Criterion and Mixed Kernel function makes the similar samples closer after projection, not similar samples farther after projection. In PIE and AR face database, the simulation results verify the effectiveness and stability of the algorithm in this paper.