In the normal goldfish, neighboring retinal ganglion cells terminate in one small tectal locus to produce the precise retinotopy characteristic of this projection. This can be directly demonstrated by labeling neighboring ganglion cells with small "spot" injections of WGA-HRP, which yield a single small patch of product at the retinotopically appropriate part of the tectum. When the optic nerve is crushed, label from these spot injections was previously found to be widely dispersed during the early phase of regeneration. With time, label subsequently condensed, typically into several discrete patches reminiscent of ocular dominance columns. In this study, we tested whether the formation of these patches required impulse activity by injecting tetrodotoxin (TTX) into the eye during regeneration. We found that impulse blockade completely inhibited the formation of discrete patches while permitting considerable condensation of the label. This implies that these patches are generated by activity but that some map "refinement" utilized cellular processes that are activity independent. This activity-independent condensation progressed at a noticeably slower rate than the equivalent condensation seen with activity, thus suggesting that activity normally participates as a "helper factor," even though it is not strictly required. Since the formation of discrete patches during regeneration provides a sensitive measure of activity-dependent refinement, this was used to further address two controversial questions concerning the role of impulse activity. One is whether there is a chronologically defined critical period for activity-dependent refinement. This was tested by blocking impulse activity for 2 to 4 months, much longer than the activity-dependent refinement is thought to last, and then permitting activity to resume. We found that multiple patches were formed following this period of late activity, thus indicating that synaptic plasticity extends for several months beyond the supposed critical period. The other question was whether spontaneous retinal activity was sufficient for activity-dependent ordering. To test this, fish were kept in constant darkness during optic nerve crush and labelled with retinal spot injections at various times during regeneration. Condensation of label with the final formation of multiple patches formed at about the same time as fish with normal visual experience. This implies that the amount and extent of correlation of spontaneous activity in retina is adequate for driving activity-dependent refinement.