In the regenerating retinotectal projection of goldfish, we have used intraocular injections of tetrodotoxin (TTX) to determine whether activity plays a role in organizing or refining the retinotopic map. Repeated injections produced a continuous 27-day block without producing extraocular effects or causing deleterious effects in the retinal ganglion cells. The retinotectal maps regenerated in the TTX fish were normally organized but the multiunit receptive fields were grossly enlarged. In control regenerates, 1-3 units (arbors of retinal ganglion cell axons) were simultaneously recorded at each penetration and their combined receptive field averaged 11-12 degrees, nearly the same as for single units. In TTX fish each penetration yielded at least 5-10 units whose receptive fields were clustered over a wider area averaging 27 degrees across. Individual ganglion cell receptive fields were assessed both by tectal and by intraretinal recording and were not enlarged. Many fish were recorded up to 4 months after the release from TTX block, but no further refinement of the maps occurred. If the nerve was recrushed and regenerated a second time without TTX, a normal map was formed, ruling out any permanent changes in the retinal ganglion cells or in the tectum. Blocks during various portions of the regeneration process showed that lack of activity during the process of axonal elongation (first 2 weeks) does not cause enlargement of the multiunit receptive fields, but lack of activity during the period of synapse formation and maturation (14-34 days) does. The results are discussed in terms of an activity-dependent stabilization of synapses. Neighboring retinal ganglion cells are known to fire in a statistically correlated fashion and this could help in their elimination of incorrect branches following an early period of diffuse connections.