This paper proposes an efficient indexing technique which can be used in an identification system with large multimodal biometric database. In this technique, multi-dimensional feature vectors of each trait (iris, signature, ear and face) are normalised and projected to a lower dimensional feature space. The reduced feature vectors are fused at feature level and used to index the database by forming Kd-tree. The performance of the proposed technique is also analysed with the feature vectors of all traits by first fusing them and projecting the fused feature vector to a lower dimensional space, and using it for indexing. Performance is also compared with the indexing based on score-level fusion. The experiment is performed on a multimodal database consisting of 5400 images of 150 subjects (i.e. nine images per subject, per trait). Out of the nine, eight images are used for training and one is used for testing. Our experiment shows that the proposed technique significantly reduces the data retrieval time along with possible error rates.