1. In Xenopus embryos, the frequency of natural and fictive swimming usually drops slowly as swimming continues but can increase following stimulation of the skin or dimming of the illumination. We have investigated whether such increases are associated with an increase in the number of neurones active at higher frequencies. 2. Recordings from ventral presumed motoneurones show that these were reliably active at all swimming frequencies. 3. Recordings from more dorsal presumed interneurones showed that in the majority of these firing probability decreased as a function of swimming frequency. Dye-filled microelectrodes were used to show that some of these neurones had the anatomy of known classes of excitatory and inhibitory premotor interneurones. 4. If skin stimulation is given at appropriate phases of the swimming cycle, it can lead to a transient increase in frequency. Recordings from silent premotor interneurones during such stimulation show that they can be recruited to fire during the post-stimulus frequency increases. 5. It was possible that spike failure in the interneurones could have been due to damage by the recording microelectrodes. We therefore measured the amplitudes and probability of occurrence of rhythmic 'on-cycle' IPSPs which occur in sensory interneurones and 'on-cycle' IPSPs which sometimes occur in motoneurones during fictive swimming. Both decreased in amplitude and could fail as frequency dropped, providing further evidence that the number of inhibitory interneurones firing on each cycle of swimming is a function of frequency. 6. We conclude that premotor rhythm-generating interneurones are not active on all cycles of swimming and that their probability of firing action potentials increases with swimming frequency. This suggests that swimming frequency is determined in part by the number of premotor interneurones which are active.