Cut optic axons regenerate into the goldfish tectum by indirect routes. To study the accuracy with which they terminate we made visuotectal maps before and immediately after cutting through the rostral tectum from its medial edge. This severed the normal pathway to medial tectum leaving intact only grossly misrouted axons. In fish mapped in this way, 3 months or more after contralateral optic nerve cuts, each initial map was essentially normal. After a tectal cut the electrical responses were usually weaker; but their receptive field positions proved to match very closely those noted previously for the same recording sites. The mean angular change was only 8.48 degrees and much of this could be accounted for by experimental errors. In fish mapped 6 months or more after attempts to cross-unite the optic tract brachia, initial maps were less regular. Some showed gross rearrangement of groups of terminals; and recognizable irregularities persisted in two fish mapped again after a further 4 or 5 months. However, maps made after tectal cuts in such fish were not noticeably less regular than the initial maps. Field position changes averaged 17.85 degrees. In both groups misrouted axons, filled with horseradish peroxidase after mapping and visualized in whole-mounts, crossed the tectum from lateral regions to reach the medial recording sites. The small field position changes in both groups confirm that axons from detectable terminals only among their retinal neighbours; nevertheless the irregular maps seen after tract cross-union lead us to expect the orderly arrangement of the normal pathway to contribute significantly to the precision of the normal map.