1. We applied morphological, pharmacological, electrophysiological, and computer simulation techniques to analyze the origin of impulse initiation in amphibian retinal ganglion cells. 2. Morphological studies of retinal ganglion cells in the mud-puppy (Necturus maculosus) and larval tiger salamander (Ambystoma tigrinum) were carried out with the use of either retrograde or intracellular labeling with horseradish peroxidase. These studies identified a characteristic thinning of the axon that begins after the initial segment of axon emerges from the ganglion cell soma or primary dendrite. Morphometric analysis of the thin segment revealed an average length of 74 μm with a standard deviation of 22 μm. For 20 conventionally placed ganglion cells, the length of the thin segment could not be correlated with soma size, initial segment diameter, or distance from the optic disk. There was also little or no correlation for seven displaced ganglion cells. The diameter of the thin segment was below reliable estimation by light microscopy. 3. We studied the possible significance of the thin axonal segment for ganglion cell impulse generation through a combination of electrophysiological recordings (intracellular and whole-cell recordings) together with computer modeling experiments. 4. Electrophysiological experiments are consistent with the idea that the thin segment and cell soma are less excitable than the initial segment region, which appears to be the principal site of initiation of the nerve impulse. The initial segment is that portion of the axon that is bounded by the soma (or proximal dendrite) at its origin and the thin segment at its distal end. 5. Computer stimulations of impulse activation were carried out with the use of two different anatomic constraints: one class of simulations did not take into account the thin segment and assumed uniform cylinder conditions, whereas the other class of simulations included a model of the thin axonal segment. These comparative simulations indicate that the thin segment must contain a relatively high density of voltage-gated Na+ channels and support impulse traffic to account for physiological observations on orthodromic and antidromic impulse propagation. In addition, to match the physiological recordings, it is necessary for both the initial segment and the soma compartments to contain moderately high levels of Na+ channels. 6. Our physiological and simulation studies are consistent with the idea that the nerve impulse is normally initiated in the initial segment of axon and then spreads to activate a somatic impulse in the retrograde direction and the axonal impulse in the anterograde direction. 7. The thin segment confines the site of impulse initiation to a relatively short portion of the axon and decreases the threshold for impulse initiation by decreasing the electrical load on the initial segment. In addition, the thin segment reduces the electrical shunt that normally lowers the input resistance of the cell. The latter mechanism can enhance the sensitivity of retinal ganglion cells. 8. An axonal site of impulse initiation may also be a feature of the retinal ganglion cells of other species.