Locally-correlated neural activity appears to play a key role in refining topographically mapped projections. The retinotectal projection of the goldfish normally regains a high degree of spatial precision after regeneration of a cut optic nerve, but it fails to do so if retinal ganglion cell activity is blocked by tetrodotoxin, or if local correlations in activity are masked by the synchronizing effect of stroboscopic light. A sharp retinal image is not normally needed for a sharp map because local correlation occurs even in darkness or diffuse light, but the possibility that a sharp image might restore local correlation and sharpen the map in stroboscopic light, though taken into account in earlier experiments, has not previously been tested. The precision of the retinotectal map was therefore studied, by retrograde transport of WGA-HRP from a standard tectal injection site and quantitative analysis of the labelled ganglion cell distribution, after regeneration of a cut optic nerve for 83-84 days in either continuous stroboscopic light or normal diurnal light. The lens of the eye was either ablated to blur the retinal image or sham-operated. Two different strobe flash patterns used in previous experiments were also compared. With the lens ablated, stroboscopic light impaired map refinement significantly, confirming previous results. A rapid, irregular flash pattern averaging about 5 Hz was rather more effective than a regular 1 Hz pattern. With the lens intact, however, neither pattern had any detectable effect. The significant gain in precision resulting from a sharp retinal image in these circumstances suggests that common mechanisms could underlie both the internal refinement of the retinotectal map and such directly experience-sensitive processes as the experimental realignment of binocular maps in the frog Xenopus, and of auditory and visual maps in the barn owl.