Delusions are bizarre and distressing beliefs that characterize certain mental illnesses. They arise without clear reasons and are remarkably persistent. Recent models of delusions, drawing on a neuroscientific understanding of learning, focus on how delusions might emerge from abnormal experience. We believe that these models can be extended to help us understand why delusions persist. We consider prediction error, the mismatch between expectancy and experience, to be central. Surprising events demand a change in our expectancies. This involves making what we have learned labile, updating and binding the memory anew: a process of memory reconsolidation. We argue that, under the influence of excessive prediction error, delusional beliefs are repeatedly reconsolidated, strengthening them so that they persist, apparently impervious to contradiction.