This article presents a theory of cortical responses based on perceptual inference and learning, emphasizing the role of hierarchical generative models and empirical Bayes. It argues that the brain minimizes free energy to infer causes of sensory inputs and learn relationships between inputs and causes. Cortical responses are seen as the brain's attempt to minimize free energy induced by stimuli, encoding the most likely cause. Learning emerges from changes in synaptic efficacy that minimize free energy, averaged over all stimuli. The theory suggests that hierarchical models allow the brain to construct prior expectations dynamically and contextually, providing a principled framework for understanding cortical organization and responses. The theory predicts that sensory cortex is hierarchically organized, with reciprocal connections and functional asymmetry between forward and backward connections. It also predicts associative plasticity and spike-timing-dependent plasticity in synaptic physiology, and accounts for classical and extra classical receptive field effects and long-latency components of evoked responses. The theory explains phenomena such as repetition suppression, mismatch negativity (MMN), and the P300 in electroencephalography, as well as psychophysical effects like priming and global precedence. It also highlights the role of perceptual learning in modifying evoked responses, as measured by MMN, and its implications for empirical studies of cortical coupling. The article discusses the anatomical and physiological basis of cortico-cortical connections, emphasizing hierarchical organization, reciprocal connections, and functional asymmetry between forward and backward connections. It also explores the role of synaptic plasticity in learning and memory, and how associative plasticity enables the brain to adapt to new information. The theory is grounded in empirical Bayes and hierarchical models, which provide a framework for understanding how the brain infers causes from sensory inputs and learns from experience. The article concludes that this theory offers a unified perspective on various aspects of cortical organization and responses, including functional specialization, integration, and plasticity.This article presents a theory of cortical responses based on perceptual inference and learning, emphasizing the role of hierarchical generative models and empirical Bayes. It argues that the brain minimizes free energy to infer causes of sensory inputs and learn relationships between inputs and causes. Cortical responses are seen as the brain's attempt to minimize free energy induced by stimuli, encoding the most likely cause. Learning emerges from changes in synaptic efficacy that minimize free energy, averaged over all stimuli. The theory suggests that hierarchical models allow the brain to construct prior expectations dynamically and contextually, providing a principled framework for understanding cortical organization and responses. The theory predicts that sensory cortex is hierarchically organized, with reciprocal connections and functional asymmetry between forward and backward connections. It also predicts associative plasticity and spike-timing-dependent plasticity in synaptic physiology, and accounts for classical and extra classical receptive field effects and long-latency components of evoked responses. The theory explains phenomena such as repetition suppression, mismatch negativity (MMN), and the P300 in electroencephalography, as well as psychophysical effects like priming and global precedence. It also highlights the role of perceptual learning in modifying evoked responses, as measured by MMN, and its implications for empirical studies of cortical coupling. The article discusses the anatomical and physiological basis of cortico-cortical connections, emphasizing hierarchical organization, reciprocal connections, and functional asymmetry between forward and backward connections. It also explores the role of synaptic plasticity in learning and memory, and how associative plasticity enables the brain to adapt to new information. The theory is grounded in empirical Bayes and hierarchical models, which provide a framework for understanding how the brain infers causes from sensory inputs and learns from experience. The article concludes that this theory offers a unified perspective on various aspects of cortical organization and responses, including functional specialization, integration, and plasticity.