Recent years, people pay more and more attention to multi-modal biometric recognition because of its higher security. In this paper, face and palmprint are fused in feature level and a novel user-specific weighting rule is used to adapt different people's need and improve recognition accuracy. According to respective characteristic of face and palmprint, different feature extraction methods are adopted. For face, phase congruency is used to extract its energy feature meanwhile Gabor transformation is used to extract texture feature of palmprint. Then the two modals are weighted according to users' specific in feature level. Experimental results show good performance.