This paper deals with Automatic Speaker Recognition in a binaural context. Such a problematic, not so widely dealt with within the speech processing community, can have potential applications in humanoid robots where speech can be used as the most natural interface between humans and robots. The proposed recognition system is based on parallel Predictive Neural Networks exploiting MFCCs (Mel Frequency Cepstral Coefficients) to discriminate multiple talkers. Because of the binaural nature of the system, the sensitivity of the proposed algorithm to the speaker spatial position during the learning step is carefully studied. The influence of noise and reverberation on the recognition rate is also reviewed. Finally, preliminary experimental results based on the recorded signals from a binaural dummy head are presented.