The retinotectal projection of the goldfish was studied after regeneration of a cut optic nerve in stroboscopic light, constant light or diurnal light, with the lens removed to blur the retinal image. Retrograde transport of wheatgerm agglutinin, conjugated to horseradish peroxidase, from a standard tectal injection site was used to measure the topographic precision of the projection. The dispersion of labelled retinal ganglion cells, which reflects this precision, was assessed by a method based on distance to nearest neighbour. In normal fish treated similarly, these cells are known to be clustered into about 1% of the retinal area. Early in regeneration, however, they are widely dispersed. The projection map then re-acquires its precision over two or three months. In diurnal light, lens ablation had no effect on refinement of the regenerated map. Constant light increased the number of labelled cells but also had no significant effect on the map. But in stroboscopic light with a continuous pseudorandom pattern of flash intervals (average rate 4.8 Hz), much less refinement was seen. Even after 70-98 days of regeneration, labelled cells remained scattered, on average, over 20% of the retinal area. These retinae were indistinguishable by several criteria from those obtained in diurnal light after only 32-39 days. Mislocated axon terminals, which are largely eliminated during the second and third months of regeneration in diurnal light, evidently persist much longer in stroboscopic light that synchronizes ganglion cell activity across the retina. These results, like previous ones obtained by blocking the transmission of activity to the tectum, support a model of map refinement based on correlation in the firing of neighbouring neurons, which may have wide application within the nervous system.