Biometric fusion is an essential procedure in any multi-modal biometric person recognition systems and it can be performed at sensor, feature, score and decision levels. This paper proposes a simulated annealing (SA) algorithm for the fusion of multi-modal biometric data. This method is applied to an Audio-Visual (AV) person recognition database that includes acoustic and visual information. Its superior performance is demonstrated through tests on 40 subjects in the AV database.