In adult goldfish, electrophysiological studies have shown that the retinotectal projection reorganizes, following removal of half of the tectum, to form a complete but compressed projection over the remaining half tectum. As a result, each fiber terminates more rostrally than normal. Electron microscopic studies suggest a competition between retinal fibers for a fixed number of synaptic sites. The current study examines whether retinal arbors in the compressed projection are smaller than normal in extent or branching and whether the fiber paths in the tectum show the rostral movements and the search strategy that the retinal fibers use. The caudal half tectum was removed without cutting retinal fibers except those at the cut edge. At 3 to 19 months afterward, retinal fibers were labeled with horseradish peroxidase. In whole-mounted tecta, fibers and terminals were drawn under camera lucida and compared with normal arbors. The axonal paths were also traced across the tectum to their termination sites. At 3 to 6 months (early stages of compression), the arbors were rather normal in appearance, although they were actually significantly larger (23%) than normal in linear extent, arborized somewhat deeper and had fewer branches (18%). The fibers normally terminating in the rostral tectum followed normal stereotyped paths, whereas those cut at the edge had grown back and forth loops (apparent searching behavior) with little branching. By 10 months when compression is complete, arbors were significantly smaller than normal (19%), were arborizing significantly deeper, and had significantly fewer branches (19%). The differences were more pronounced in arbors of coarse and medium caliber than in fine caliber axons. The axons still ran in stereotyped fascicles, but included an extrafascicular portion that, unlike any axons in normals, turned back in a rostral direction before branching. This striking effect, present even in far rostral tectum, indicated that arbors had been forced to move rostrally to accomodate those from the ablated half. The small effect on arbor extent suggests that this is influenced by factors other than the magnification factor of the map, perhaps postsynaptic dendritic extent. The increased depth of termination is consistent with the increased thickness of the retinal terminal layer. The decreased number of branches is consistent with the conclusion that the remaining fixed number of synaptic sites shared among the full complement of retinal fibers should result in fewer synapses per retinal fiber.