The goal of this work is the improvement of the performance of a multimodal biometric identification system based on fingerprints and finger vein recognition. This system has to authenticate the person identity using features extracted from his fingerprints and finger veins by multimodal fusion. It is already proved that multimodal fusion improves the performance of biometric recognition, basically the fusion at feature level and score level. However, both of them showed some limits and in order to enhance the overall performance, a new fusion method has been proposed in this work; it consists on using both features and scores fusion at the same time. The main contribution of investigation this technique of fusion is the reduction of the template size after fusion without influencing the overall performance of recognition. Experiments were performed on a real multimodal database SDUMLA-HMT and obtained results showed that as expected multimodal fusion strategies achieved the best performances versus uni-modal ones, and the fusion at feature level was better than fusion at score level in recognition rate (100%, 95.54% respectively) but using more amounts of data for identification. The proposed hybrid fusion strategy has overcome this limit and clearly preserved the best performance (100% as recognition rate) and in the same time it has reduced the proportion of essential data necessary for identification.