The prevailing model of amnesia holds that it depends on damage to a group of loosely related structures, collectively called the medial temporal lobe (MTL). One influential theory holds that these areas work together as a single functional unit. According to this theory, the structures in the MTL store the memory of specific objects, facts and events. We have developed a competing hypothesis, which holds that different structures within the MTL have specialized functions, and that their collective functions extend beyond those assigned to the MTL by the prevailing model. To test our hypotheses, we have made use of a task that measures the rate of acquisition of arbitrary associations between visual stimuli and spatially-directed responses, known as visuomotor learning (VML). Amnesic patients are notoriously poor in acquiring arbitrary associations, but there is little precise knowledge of the brain areas underlying VML in healthy subjects. Our past work has shown that transection of the fornix, like aspirations lesions of the hippocampal formation plus underlying cortex, disrupts VML. In the current review period, we determined the precise part of the hippocampal system that is critical for VML. Our research during the current review period has examined the contributions of specific brain regions that have undergone temporary inactivation. This approach provides within-subject measures of memory that can be more sensitive than comparisons across groups. Subjects rapidly acquired arbitrary associations between visual stimuli and spatial goals. We then assessed the effects of reversible inactivations of either the hippocampus or the entorhinal cortex on the within-session rate of learning. For comparison, we also evaluated the effects of inactivation on the performance of previously learned associations between visual stimuli and goals. These associations had been acquired long before the inactivations, were very familiar to the subject, and were performed nearly flawlessly. We found that inactivation of the entorhinal cortex disrupted acquisition of arbitrary visuomotor associations but left intact performance of the previously learned associations of the same type. In contrast, inactivation of the hippocampus had no effect on either new learning or performance according to the familiar associations. These data indicate that the entorhinal cortex, but not the hippocampus, is causally involved in establishing new visuospatial associations, as opposed to retrieving previously learned associations. Notably, this is the same pattern of impairment seen after fornix transection, suggesting that the entorhinal cortex projections into the fornix, or perhaps via the subiculum, are critical for VML and, by extension, the fast learning of arbitrary associations generally.