1. Simulations of Xenopus embryo spinal neurons were endowed with Hodgkin-Huxley-style models of voltage-dependent Na+, Ca2+, slow K+ and fast K+ currents together with a Na(+)-dependent K+ current. The parameters describing the activation, inactivation and relaxation of these currents were derived from previous voltage-clamp studies of Xenopus embryo spinal neurons. Each of the currents was present at realistic densities. 2. The model neurons fired repetitively in response to current injection. The Ca2+ current was essential for repetitive firing in response to current injection. The fast K+ current appeared mainly to control spike width, whereas the slow K+ current exerted a powerful influence on the reptitive firing properties of the neurons without markedly affecting spike width. 3. The properties of the model neurons could be made more consistent with those previously reported for Xenopus embryo neurons during intracellular recordings in vivo, if the shunting effect of the sharp microelectrode was incorporated into the model. 4. The model neurons were then used to create a simplified version of the spinal network that controls swimming in the frog embryo. This model network could generate the motor pattern for swimming: the activity between the left and right sides alternated with a cycle period that varied from 50 to 120 ms. This is very similar to the range of cycle periods observed in the real embryo. The shunting effect of the microelectrode was once again taken into account. 5. Reductions of the K+ currents perturbed the motor pattern and gave three forms of aberrant motor activity very similar to those previously seen during the application of K+ channel blockers to the real embryo. The ability to generate the correct motor pattern for swimming in the model depended on the balance between the K+ currents and the inward Na+ and Ca2+ currents rather than their absolute values. 6. The model network could generate a motor pattern for swimming over a very wide range of excitatory (2-10 nS) and inhibitory (2-400 nS) synaptic strengths. Rough estimates of the physiological synaptic strengths in the real circuit (around 20-60 nS for inhibition and 2-5 nS for excitation) fall within the range of synaptic strengths that gave simulation of the swimming motor pattern in the model. 7. The cycle period of the motor activity in the model shortened either as the excitatory synapses were strengthened or as the inhibitory synapses were weakened. 8. The prediction that the strength of the mid-cycle inhibition determines cycle period has been tested by using low levels of strychnine to reduce glycinergic reciprocal inhibition in a graded manner in the real embryo. As the inhibition was reduced, the cycle period of fictive swimming in the embryo shortened by amounts very close to those predicted by the model. 9. This new experimentally derived model can replicate many of the known features of fictive swimming in the real embryo and may be of value as an analytical tool in attempting to understand how the spinal circuitry of the Xenopus embryo and related amphibian embryos control a variety of motor behaviours.