Patients suffering from schizophrenia are less susceptible to various visual illusions. For example, healthy participants perceive a hollow mask as a normal face, presumably due to the strength of constraining top-down influences, while patients with schizophrenia do not (Schneider, U., Leweke, F.M., Sternemann, U., Weber, M.M., Emrich, H.M., 1996. Visual 3D illusion: a systems-theoretical approach to psychosis. Eur. Arch. Psychiatry Clin. Neurosci. 246, 256-260; Scheider, U., Borsutzky, M., Seifert, J., Leweke, F.M., Huber, T.J., Rollnik J.D., Emrich, H.M., 2002. Reduced binocular depth inversion in schizophrenic patients. Schizophrenia Research 53, 101-108.; Emrich, H.M., Leweke, F.M., Schneider, U., 1997. Towards a cannabinoid hypothesis of schizophrenia: cognitive impairments due to a dysregulation of the endogenous cannabinoid system. Pharmacol. Biochem. Behav. 56, 803-807). However the neural mechanisms underpinning this effect remain poorly understood. We used functional magnetic resonance imaging to investigate the hollow-mask illusion in schizophrenic patients and healthy controls. The primary aim of this study was to use measures of effective connectivity arising from dynamic causal modelling (DCM) to explain differences in both the perception of the hollow-mask illusion and associated differences in neural responses between patients with schizophrenia and controls, which we hypothesised would be associated with difference in the influences of top-down and bottom-up processes between the groups. Consistent with this explanation, we identified differences between the two groups in effective connectivity. In particular, there was a strengthening of bottom-up processes, and weakening of top-down ones, during the presentation of 'hollow' faces for the patients. In contrast, the controls exhibited a strengthening of top-down processes when perceiving the same stimuli. These findings suggest that schizophrenic patients rely on stimulus-driven processing and are less able to employ conceptually-driven top-down strategies during perception, where incoming sensory data are constrained with reference to a generative model that entails stored information from past experience.