Normalization is an essential step in multimodal biometric system that involves various nature and scale of outputs from different modalities before employing any fusion techniques. This paper proposes score normalization technique based on mapping function to increase the separation of score at overlap region and reduce the effect of overlap region on fusion algorithm. The effect of the proposed normalization technique on recognition system performance for different fusion methods is examined. Experiments on three different NIST databases suggest that integrating the proposed normalization technique with the classical simple rule fusion strategies (sum, min and max) and SVM-based fusion results significant improvement compared to other baseline normalization techniques used in this work.