This paper explores the use of previously described computational model of sections of the mammalian auditory system (Pont and Damper, 1991) as the "front end" to a speech recognition system based on a conventional neural network. In the paper, experiments using two different auditory front ends were implemented, the first based on a model of the auditory nerve (AN), the second based on a model of the dorsal cochlear nucleus (DCN). In each case, the neural network was a multi-layered Perceptron. In the first experiment, the two hybrid recognisers were tested on isolated digits recorded in quiet conditions. Here it was found that both AN- and DCN-based models performed excellently. In the second experiment the same stimuli were used, but this time with added noise. In this case it was found that the performance of the AN-based recogniser remained high, but - as the SNR was decreased - the performance of the DCN-based device fell of dramatically. The results are discussed and some suggests for further studies are made.