This article aims to provide a theoretical framework to elucidate the neurophysiological underpinnings of deviance detection as reflected by mismatch negativity. A six-step model of the information processing necessary for deviance detection is proposed. In this model, predictive coding of learned regularities is realized by means of long-term potentiation with a crucial role for NMDA receptors. Mismatch negativity occurs at the last stage of the model, reflecting the increase in free energy associated with the switching on of silent synapses and the formation of new neural circuits required for adaptation to the environmental deviance. The model is discussed with regard to the pathological states most studied in relation to mismatch negativity: alcohol intoxication, alcohol withdrawal, and schizophrenia. [This article is produced by Elsevier, Orlando, FL.]