The buccal ganglia of Aplysia contain a central pattern generator (CPG) that organizes the rhythmic movements of the radula and buccal mass during feeding. Many of the cellular and synaptic elements of this CPG have been identified and characterized. However, the roles that specific cellular and synaptic properties play in generating patterns of activity are not well understood. To examine these issues, the present study developed computational models of a portion of this CPG and used simulations to investigate processes underlying the initiation of patterned activity. Simulations were done with the SNNAP software package. The simulated network contained two neurons, B31/B32 and B63. The development of the model was guided and constrained by the available current-clamp data that describe the properties of these two protraction-phase interneurons B31/B32 and B63, which are coupled via electrical and chemical synapses. Several configurations of the model were examined. In one configuration, a fast excitatory postsynaptic potential (EPSP) from B63 to B31/B32 was implemented in combination with an endogenous plateau-like potential in B31/B32. In a second configuration, the excitatory synaptic connection from B63 to B31/B32 produced both fast and slow EPSPs in B31/B32 and the plateau-like potential was removed from B31/B32. Simulations indicated that the former configuration (i.e., electrical and fast chemical coupling in combination with a plateau-like potential) gave rise to a circuit that was robust to changes in parameter values and stochastic fluctuations, that closely mimicked empirical observations, and that was extremely sensitive to inputs controlling the onset of a burst. The coupling between the two simulated neurons served to amplify exogenous depolarizations via a positive feedback loop and the subthreshold activation of the plateau-like potential. Once a burst was initiated, the circuit produced the program in an all-or-none fashion. The slow kinetics of the simulated plateau-like potential played important roles in both initiating and maintaining the burst activity. Thus the present study identified cellular and network properties that contribute to the ability of the simulated network to integrate information over an extended period before a decision is made to initiate a burst of activity and suggests that similar mechanisms may operate in the buccal ganglia in initiating feeding movements.