Prediction is at the core of memory. A memory system stores information about the past in service of preparation for the future, and thus the act of memory retrieval may be viewed as an act of prediction about upcoming events. Converging evidence from animal and compuational models suggests that within the hippocampus, stored memories are compared to current sensory input in order to compute novelty -- when expectation deviates from actual outcome. Specifically, hippocampal subfield CA1 is thought to support this computation of mismatch between past and present. Detection of novelty in turn is hypothesized to modulate encoding processes, providing a mechanism for gating the entry of information into long-term memory. Using high-resolution functional MRI, I examined hippocampal subfield (CA1, CA23/DG[dentate gyrus], subiculum) and medial temporal lobe (MTL: entorhinal, perirhinal, parahippocampal) cortical activation during associative retrieval and associative mismatches in humans. In Experiment 1, subjects performed cued image retrieval and made explicit comparisons of memory to matching or mismatching decision probes. Activity in multiple hippocampal and MTL cortical subregions tracked associative retrieval success, whereas activity in CA1 and perirhinal cortex tracked the presence of associative mismatches. In Experiment 2, subjects viewed sequences of images while performing an incidental task (1-back target detection). In CA1, CA23/DG, and perirhinal cortex, activation was greater when image sequences were presented in rearranged order (mismatch with memory) compared to repeated order (match with memory). In CA1 only, this mismatch enhancement was significantly modulated by prediction strength: the mismatch enhancement was greater when predictions were stronger. In a separate behavioral experiment, recognition memory was found to be better for images that had appeared in rearranged-order (mismatch) than repeated-order (match) sequences, supporting the notion that mismatch detection leads to encoding upregulation. In Experiment 3, subjects viewed sequences of images containing either temporal predicitive information only or combined temporal and spatial predictive information. When novel images were embedded within previously viewed sequences, activation in CA1 was significantly related to subsequent memory for items which had violated predictions (remembered> forgotten), with this enhancement modulated by the amount of predictive information. Together, these data are consistent with the hypothesis that CA1 acts as a comparator, detecting when memory for the past and sensory input in the present diverge. More broadly, the current studies reveal the dynamics of the human hippocampus and MTL cortices during the acts of prediction, novelty detection, and memory encoding -- a continuous cycle of events that enable preparation for the future based on past experience.