Intracellular recordings were made from the neurites of interneurons and motoneurons in the metathoracic ganglion of the cockroach, Periplaneta americana. Many neurons were penetrated which failed to produce action potentials on the application of large depolarizing currents. Nevertheless, some of them strongly excited and/or inhibited slow motoneurons innervating leg musculature, even with weak depolariziing musculature, even with weak depolarizing currents. Cobalt-sulfide-straining of these nonspiking neurons showed them to be interneurons with their neurites contained entirely within the metathoracic ganglion. Two further characteristics of these interneurons were rapid spontaneous fluctuations in membrane potential and a low resting membrane potential. One nonspiking neuron, interneuron I, when depolarized caused a strong excitation of the set of slow levator motoneurons which discharge in bursts during stepping movements of the metathoracic leg. During rhythmic leg movements the membrane potential of interneuron I oscillated with the depolarizing phases occurring at the same time as bursts of activity in the levator motorneurons. No spiking or any other nonspiking neuron was penetrated which could excite these levator motoneurons. From all these observations we conclude that oscillations in the membrane potential of interneuron I are entirely responsible for producing the levator bursts, and thus for producing stepping movements in a walking animal. During rhythmic leg movements, bursts of activity in levator and depressor motoneurons are initiated by slow graded depolarizations. The similarity of the synaptic activity in these two types of motoneurons suggests that burst activity in the depressor motoneurons is also produced by rhythmic activity in nonspiking interneurons. The fact that no spiking neuron was found to excite the depressor motoneurons supports this conclusion. Interneuron I is also an element of the rhythm-generating system, since short depolarizing pulses applied to it during rhythmic activity could reset the thythm. Long-duration current pulses applied to interneuron I in a quiescent animal did not produce rhythmic activity. This observation, together with the finding that during rhythmic activity the slow depolarizations in interneuron I are usually terminated by IPSPs, suggests that interneuron I alone does not generate the rhythm. No spiking interneurons have yet been enccountered which influence the activity in levator motoneurons. Thus, we conclude that the rhythm is generated in a network of nonspiking interneurons. The cellular mechanisms for generating the oscillations in this network are unknown. Continued.