In this paper, a novel speaker-identification (speaker-ID) technique based on explicit exploitation of stochastic characteristics of test utterance is proposed. Unlike the conventional approach which hypothesizes the identity of a test speaker by determining which client's model maximizes the likelihood for the test utterance, it is aimed to bilaterally compare test speaker's voices with client speakers' voices instead of simply taking the unilateral likelihoods into account. We study two approaches respectively based on cross likelihood ratio and Bayesian information criterion to accomplish this aim. Performance of the proposed approaches was evaluated by close-set text-independent speaker ID experiments and was shown to be superior to that of the conventional approach based on maximum likelihood decision rule.