The overarching objective of this research program is to use neuroimaging methods to determine how the recent past is coded in the human brain and how this coding contributes to the processing of incoming information. A central tenet of this proposal is that being able to maintain a representation of the recent past is fundamental for constructing internal predictions about future states of the environment. The construction of such has been called predictive coding, such predictions have been argued to play a fundamental role in disambiguating signal information from a noisy or degraded array.We implement a comprehensive and multi-disciplinary research program to understand how regularities in the recent past are coded, and how they give rise to predictive codes of future states. On the basis of prior work we propose that disambiguation of signals is performed by a predictive system that relies strongly on representing the statistical properties of the recent past. This system is instantiated via interactions between three neural systems: (1) medial temporal structures including the hippocampus and parahippocampal cortex that encode statistical features of the recent past and signal whether predictions are licensed, (2) higher level cortical regions that code for detailed predictions in various modalities and generate efferent top-down predictions, and (3) lower-level sensory cortices whose activity at any given moment reflects not only bottom-up processing of sensory inputs, but also the assessment of these inputs against top-down predictions propagated from higher-levels regions. We will use neuroimaging methods with high spatial and temporal resolution (fMRI, MEG) to study neural activity in these three neural systems and the interaction between them.