The predictive coding model proposed by Friston (2005) makes very specific prediction for the behavior of ERP recorded in oddball tasks, which are informed by the biological constraints inherent in hierarchical sensory systems. I will briefly review evidence for and against the following predictions: 1) existence of hierarchical MMN generators, 2) dependence on the probabilistic structure of the stimulus sequence, 3) the existence of repetition (learning) effects to standards, 4) involvement of NMDA receptor-dependent short-term plasticity and 5) modulation by acetylcholine receptors in learning-related changes. 6) Methods for evaluating functional interactions between hierarchical auditory areas during repetition-related learning will be reviewed for humans by Garrido et al. (same symposium) and in animals by Malmierca et al. (same meeting). I will conclude with suggestions for how this model could inform further investigations of developmental language disorders (e.g. dyslexia) and psychiatric disorders (e.g. schizophrenia).