This experiment used a 17-arm radial maze, but only 8 arms were ever baited with food. For one group the baited arms were adjacent, and for the other, the “mixed” group, baited arms were interspersed with unbaited arms. Both groups learned after about 30 days of testing to perform accurately, rarely going down an unbaited arm and having a low probability of revisiting a baited arm on the same day.
Following this training, half the animals in each group received lesions to the fimbria (part of the hippocampus) and fornix (the output pathway for the hippocampus).
The animal with large lesions of this type still rarely made errors of going down an unbaited arm, but quite often revisited baited arms on the same day. These revisits are described by the authors as errors of working memory — "The results of this experiment demonstrate differential involvement of the hippocampal system in reference and working memory".