Hippocampectomized and sham-operated male rats were pre-or postoperatively trained to perform a reference/working-memory task in a standard radial eight-arm maze. On each trial, three arms were baited. These "correct" arms formed one of two configurations, which changed between conditions. The remaining arms were never baited. The rats' task was to enter each of the baited arms only once, and to refrain from entering any of the unbaited arms. Incorrect choices were punished with a brief period of confinement. Preoperatively trained groups were tested postoperatively for retention, and then trained on a new reference-memory task in which the baited arms formed the second configuration. Postoperatively trained groups were also trained on two such tasks in succession. Order of training on the two configurations was counterbalanced between groups. Postoperatively trained hippocampal groups were severely impaired in acquisition of the tasks. Preoperatively trained hippocampal groups were unimpaired, relative to shamoperated groups, in all measures of performance.