We have monitored the spectrum of the (spontaneous) neural noise at the round window (RW) and on the surface of the antero-ventral cochlear nucleus (CN) and the dorsal CN (DCN) of anaesthetised guinea pigs. We have also obtained the average gross extracellular waveform evoked by 20 kHz tone-bursts (0.25 ms and 25 ms) at each of these recording sites, and calculated the spectrum of the average waveforms (SAW). With these tone-bursts, only a small population of neurones in the extreme basal turn of the cochlea near the RW electrode responds, presumably with only a single action potential for each 0.25 ms tone-burst. The RW waveforms recorded between 20 dB and 60 dB SPL were very similar, and are therefore presumably a simple estimate of the shape of the contribution of the firing of a single neurone to the gross RW signal (the unitary potential or UP). In normal animals, the SNN and the SAW were remarkably similar, with peaks at 900 Hz and at 2400 Hz, suggesting that they are not due to neural synchronisation (as suggested previously by others), but are due to an oscillatory waveform produced by each single fibre action potential. Abolition of all spike activity by RW tetrodotoxin left a waveform with only a summating potential and a dendritic potential, and no 900 Hz peak in the SAW or SNN, indicating that the spectral peak is due to neural spiking only. Abolition of the CN contribution to the RW waveforms by CN application of lignocaine or sectioning of the cochlear nerve at the internal meatus (by focal aspiration of the DCN and underlying cochlear nerve) showed that the 900 Hz peak was not simply due to the addition of a delayed and inverted CN contribution: mathematical modelling shows that this would produce a broad spectral peak at about 1200 Hz. Moreover, the 900 Hz spectral peak remains after complete abolition of the CN contribution, although reduced in amplitude. This residual 900 Hz peak can be traced to an oscillation in the gross waveform due to the presence of two peaks (P(1)* and N(2)*) which follow the intact N(1) peak. The P(1)* and N(2)* peaks were present at the RW, but not at the cochlear nerve as it exits the internal meatus, suggesting that they were not due to double-spiking of some of the neurones, but were probably due to a sub-threshold electrical resonance in the peripheral dendrites. We have successfully modelled the production of the SNN and the compound action potential and SAW in response to 0.25 ms and 25 ms tone-bursts at 20 kHz by including only a damped 900 Hz resonance in the UP, without refractory effects, preferred intervals or synchronisation in the timing of neural spike generation. Such resonances in other neurones are known to be due to the activation kinetics of the voltage-controlled sodium (Na(+)) channels of these neurones. The presence of such sub-threshold oscillations probably indicates that the peripheral dendrites are devoid of stabilising potassium (K(+)) channels. We also discuss the role of this membrane resonance in generating burst-firing of the cochlear nerve (as with salicylate) and the role of such burst-firing in generating tinnitus.