This paper deals with a new phoneme recognition system based on a model of human auditory system. This system is made up of a model of human cochlea and a simple multi-layer recurrent neural network which has feedback connections of self-loop type. The ability of this system has been investigated by a phoneme recognition experiment using a number of Japanese words uttered by a native male speaker. The result of the experiment shows that recognition accuracies achieved with this system in the presence of noise are higher than those obtained by a combination of frequency spectral analysis by DFT and conventional feedforward neural network and that the cochlea model effectively prevents the deterioration due to noise of recognition accuracy.