Abstract: A new speaker recognition method, based on the D-S evidence theory, is proposed. D-S evidence theory is a useful method to deal with uncertainty and can combine evidences from different evidence sources. Therefore it has been successfully applied to data fusion and objection recognition. In this paper, features of the speakers are extreacted and fused with D-S approach, the basic probability assignment are assigned repeatedly. As a result, the reliability is obtained and the identification of speakers can be achieved efficiently. The experiments indicate that speaker recognition based on Dempster-Shafter evidence theory has better effect than speaker recognition based on vector quantization.