In recent years, biometric-based authentication systems have become very important in view of their ability to prevent identity theft by identifying an individual with high accuracy and reliability. Multimodal biometric systems have now drawn some attention in view of their ability to provide a performance superior to that provided by the corresponding unimodal biometric systems by utilizing more than one biometric modality. The existing multimodal biometric systems fuse multiple modalities at a single level, such as sensor, feature, score, rank or decision, and no study to fuse the modalities at more than one level that may lead to a further improvement in the performance of multimodal biometric systems, has been hitherto undertaken. In this thesis, multimodal biometric systems, wherein fusions of the modalities are carried out at more than one level, are investigated. In order to improve the performance of multimodal biometric systems over unimodal biometric systems, normalization and weighting of scores from multiple matchers are essential tasks. In view of this, in the first part of the thesis, a number of normalization and weighting techniques under the score level fusion are investigated. Unlike the existing normalization techniques that are based only on the genuine scores, four new techniques based on both the genuine and impostor scores, are proposed. Two weighting techniques that are based on confidence of the scores, are proposed. Extensive experiments are conducted to evaluate the performance of the multimodal biometric system under the score-level fusion (MBS-SL) using the proposed normalization and weighting techniques. The focus of the second part of this thesis is on the development of multimodal biometric systems, wherein fusions of the modalities are carried out at multiple levels. Specifically, two multimodal biometric systems, in which three modalities are used for their fusion both at the feature level and the score level, are proposed. In the first multimodal biometric system, referred to as the multimodal biometric system with feature level and score level (MBS-FSL) fusions, the features of the three modalities are encoded using the binary hash encoding technique. Unlike the existing techniques for feature level fusion that use unencoded features, this encoding technique allows the neighbourhood feature information to be taken into account. The score-level fusion is carried out on the score obtained from the feature-level fusion and the score from the matching module of the modality that has the lowest equal error rate. In the proposed MBS-FSL, the border values of raw features could not participate in the encoding in view 4-connected neighbors not being available. In order to take both the border and non-border information as well as the neighbourhood information into consideration, a second multimodal biometric system, referred to as the multimodal biometric system with modified feature level and score level (MBS-MFSL) fusions, is proposed, wherein both the raw and encoded features are taken into account. In this system, the feature-level fusion is carried out in a manner similar to that for the MBS-FSL system. The score-level fusion is then carried out between the score obtained from the feature-level fusion, the score from the matching module of the modality that was not utilized in the feature-level fusion, and the scores from individual modalities by using their raw features. Extensive experiments are performed to evaluate the performance of the two proposed multimodal biometric systems. The results of these experiments demonstrate that both of the proposed multimodal biometric systems provide performance superior to that provided by the existing multimodal biometric systems in which fusion of modalities is carried out at a single level, namely, the score level. Experimental results also show that, in view of both the border and neighbourhood feature information being considered in the proposed MBS-MFSL system, it provides a performance superior to that provided by MBS-FSL system. The investigation undertaken in this thesis is aimed at advancing the present knowledge in the field of human biometric identification by considering, for the first time, the fusion of the modalities at two levels, namely, the feature and score levels, and it is hoped that the findings of this study would pave the way for further research in the development of new multimodal biometric systems employing fusion of modalities at multiple levels.