The faithful production of rhythms by many neural circuits depends critically on the strengths of inhibitory synaptic connections. We propose a model in which the strengths of inhibitory synapses in a central pattern-generating circuit are subject to activity-dependent plasticity. The strength of each synapse is modified as a function of the global activity of the postsynaptic neuron and by correlated activity of the pre- and postsynaptic neurons. This allows the self-assembly, from random initial synaptic strengths, of two cells into reciprocal oscillation and three cells into a rhythmic triphasic motor pattern. This self-assembly illustrates that complex oscillatory circuits that depend on multiple inhibitory synaptic connections can be tuned via simple activity-dependent rules.