Learning associations between co-occurring events enables us to extract structure from our environment. However, the neuronal processes underlying the formation of new associations in the human brain are not yet well understood. Human participants, implanted with depth electrodes, learned arbitrary associations between images presented in an ordered, predictable sequence. During learning we recorded from medial temporal lobe (MTL) neurons and found that single MTL neurons showed asymmetric shifts in activity and started firing earlier in the sequence in anticipation of their preferred stimulus. In contrast, neurons in temporal cortex did not show anticipation effects as a result of this short-term learning. To investigate the consequences of longer-term associative learning, we performed an fMRI study in which subjects learned arbitrary new associations between pairs of unrelated objects over 15 days. We found that as a result of this long-term learning, multi-voxel representations of associated objects in the ventral visual pathway, but not the MTL, became more similar to each other. These results demonstrate flexible representations in the MTL that could support rapid learning, while longer-term learning could be reflected in the activity of visual regions.