To restore meaningful vision to blind patients, individual prosthetic electrodes must reliably elicit specific spiking patterns in ganglion cells. Here we have developed a stimulus protocol that replicates normal light-elicited spike patterns. On-cell patch clamp recordings were used to measure spiking responses from individual retinal ganglion cells in the flat mount rabbit retina. Small tipped platinum-iridium epiretinal electrodes were used to deliver biphasic electrical stimulus pulses; stimulation frequencies ranged from 1-250 Hz. To distinguish spiking elicited by direct activation of the ganglion cell from spiking elicited by activation of presynaptic cells, synaptic inputs to ganglion cells were blocked pharmacologically. Light responses and dendritic morphology were used to identify each ganglion cell type. Long duration electrical pulses (1 msec) elicited a single spike within 0.5 msec of the pulse onset followed by a train of spikes which could persist for more than 50 ms depending on pulse amplitude levels. Pharmacological blockers of excitatory synaptic input eliminated all but the first spike suggesting that the first spike arises from direct activation of the ganglion cell and all other spikes arise from depolarization generated by activation of excitatory presynaptic cells. Single, short duration pulses (0.1 msec) elicited a single directly-driven spike but did not generate spikes arising from presynaptic activity. Repetitive short duration pulses each reliably elicited only a single (direct) spike at all frequencies tested (<=250 Hz). Short duration pulses elicit one spike per pulse at all stimulation frequencies; therefore they can be used to generate precise temporal patterns of activity in ganglion cells. These patterns can be used to mimic physiologically relevant light evoked responses, e.g., they can replicate the spiking pattern of transient or sustained cells, and also mimic the changes in spike frequency that underlie responses to changing intensities and contrasts.