Our aim was to test the hypothesis that the frequency of neuronal rhythm-generating networks is partly controlled by the size of the active premotor interneuron population. We have tested possible mechanisms for frequency changes in a population model of the Xenopus laevis embryo spinal rhythm-generating networks for swimming. After initiation by a brief sensory excitation, the frequency of swimming activity decreases to a steady level determined by the properties of the 24 interneurons and their connections. The initial frequency decrease was dependent on the time-course of initiating sensory synaptic excitation. When some premotor excitatory interneurons were given weaker synaptic connections to reflect the variability in the spinal cord, they could drop out and stop firing during the initial frequency decrease while swimming activity continued. If the synaptic input of such weak excitatory interneurons was graded finely, they could drop out consecutively. This led to further decreases in the level of tonic excitation and in network frequency which depended on the number, type and distribution of excitatory interneurons that stopped firing. Silent weak excitatory interneurons could be recruited by a second sensory excitation and cause an increase in tonic depolarization and frequency which outlasted the sensory input. Such recruitment could occur on both sides after local sensory stimulation to only one region or one side of the model. We conclude that these computer simulations support the hypothesis that premotor interneuron drop-out and recruitment is one mechanism which can control frequency in a locomotor central pattern generator.