Hallucinations, percepts with out external stimulus, can be a very distressing feature of schizophreniform psychotic illnesses. They are trypically with D2 dopamine receptor blocking drugs. However up to 30% of patients may experiences residual hallucinations despite adequate dosing and those that do are at higher risk of harm to themselves and others. We propose to address this unmet clinical need through computational psychiatry. We have devised a task that can engender hallucinations in the laboratory. Our work thus far suggests that individuals prone to hallucinations in their everyday lives are signioficantly more susceptible to the lab effect. Our computational analyses of participant behavior reveal that people who hear voices in the context of a psychotic illness do so because they develop strong perceptual beliefs and those beliefs are resistant to updating. In our brain imaging work, perceptual beliefs are associated with insula hyperactivity and poor belief updating is associated with a dearth of cerebellar activity. We would like to develop these observations into biomarkers to better guide individuals toward particular treatment approaches and to design improved treatments that target underlying deficits. Here we propose the first step in that process; direct, causal, hypothetico-deductive, tests of the relationships between computational processes, brain activity and task behavior. We will characterize the impact of transcranial magnetic stimulation (compared to sham stimulation) on the perfomance of hallucinating patients on our lab-based hallucination task. Two specific aims are proposed: Specific Aim 1 will target the link between insula responses and strong prior beliefs. We predict that by inactivating the insula with inhibitory TMS we will reduce the patients’ proclivity towards task induced hallucinations. Specific Aim 2 will address the role of the cerebellum in belief updating. By boosting cerebellar engagement with theta-burst stimulation, we aim to augment the ability of hallucinating patients to detect changes in the task contingencies and update their priors accordingly. Each aim will be accompanied by control measures that will confirm target engagement. We believe that priors are driving the experience of hallucinations in our task, and outside of the lab. If we are correct, the data that we gather will be the first step towards task and computational model-based development of new treatments, as well as diversion of patients towards those treatmenst based on computational parameters. However, even if we are wrong, we will learn that the strong prior model is lacking and that we should pursue other explanations for hallucinations.