We have measured the compound action potential (CAP) evoked by very brief high-frequency tone-bursts (20 kHz, 1/4 ms) at the round window (RW) and at the surface of the cochlear nucleus (CN) in guinea pigs before, during and after a localised chilling of either the cochlea or CN, with a non-toxic 'freeze spray'. CN chilling almost abolished the negative-going component of the CAP measured in the CN (generated by the CN and here called the cochlear nucleus response or CNR), leaving a positive-going localised response from the cochlear neurones as they leave the internal auditory meatus. Within 3 min, the CNR recovered to control values. During that time, the N(1) component of the RW CAP was slightly increased and the P(1) was larger, even though the CNR was abolished, indicating that the P(1) was not due to electrotonic spread of current from the CN. The N(2) and successive peaks at the RW were also abolished, but returned after 30 s. When the cochlea was chilled, the RW CAP was initially reduced in amplitude, presumably due to a drop in the number of cochlear neurones spiking in response to sound, but recovered within 3 min to be larger than the control waveform, with a more prominent N(1) peak which was delayed slightly, making the CAP more monophasic. At the same time, the CNR was smaller, presumably due to fewer cochlear neurones responding, but overall the CN CAP was altered little in waveshape. These experiments indicate that that RW CAP is generated almost solely by cochlear neurones. We also suggest that some of the changes in the RW CAP during the chills were due to changes in the firing of the lateral olivo-cochlear system of efferent neurons.