This dissertation focuses on the biological structures that allow animals to exhibit classical conditioning. The project presents a computational neuroscience model combining insights from two influential accounts of learning. The first, Chorley & Seth (2011), offers a biologically realistic model of dopamine activity and association learning. The second, Redish et al. (2007), is an abstract model of internal state representation, perhaps residing in the hippocampus, that accounts for many classical conditioning phenomena. Combining these two models produces a biologically realistic explanation of both association learning and unlearning. The proposed model exhibits classical conditioning phenomena while demonstrating how spiking neurons in the brain could implement reward prediction. Specifically, this dissertation project makes the following original contributions: 1) a replication of the model of Chorley and Seth (2011), 2) the presentation of a new model, based on Chorley and Seth (2011) but incorporating a simple spiking neuron model of the hippocampus inspired by Redish et al, (2007), 3) a demonstration that the hippocampal model continues to exhibit association learning, 4) and now exhibits extinction, 5) reinstatement, 6) spontaneous recovery, 7) and blocking, 8) as well as a demonstration that the hippocampal model mirrors the results of relevant lesion studies, 9) and, finally, an analysis of how the hippocampus model represents context. While previous models of learning and unlearning based on temporal difference methods are powerful and account for some classical conditioning phenomena, this dissertation suggests that, contrary to temporal difference methods, learning and unlearning associations are not the result of a single mechanism.