39 goldfish, either normal, sham operated, or forebrain ablated, were tested in the acquisition, extinction, and retention of an instrumental avoidance response with either a short or long delay between operations and testing. Forebrain removal severely impaired or prevented acquisition, greatly reduced resistance to extinction, and completely abolished a previously learned response. The deficits were permanent-an 8-wk. delay between operations and testing did not result in improved performances. Median escape latencies did not differ between groups, indicating that the deficit is specific to avoidance learning.