Visual hallucinations are frequent, disabling complications of advanced Parkinson's disease, but their neuroanatomical basis is incompletely understood. Previous structural brain magnetic resonance imaging studies suggest volume loss in the mesial temporal lobe and limbic regions in subjects with Parkinson's disease with visual hallucinations, relative to those without visual hallucinations. However, these studies have not always controlled for the presence of cognitive impairment or dementia, which are common co-morbidities of hallucinations in Parkinson's disease and whose neuroanatomical substrates may involve mesial temporal lobe and limbic regions. Therefore, we used structural magnetic resonance imaging to examine grey matter atrophy patterns associated with visual hallucinations, comparing Parkinson's disease hallucinators to Parkinson's disease non-hallucinators of comparable cognitive function. We studied 50 subjects with Parkinson's disease: 25 classified as current and chronic visual hallucinators and 25 as non-hallucinators, who were matched for cognitive status (demented or non-demented) and age (± 3 years). Subjects underwent (i) clinical evaluations; and (ii) brain MRI scans analysed using whole-brain voxel-based morphometry techniques. Clinically, the Parkinson's disease hallucinators did not differ in their cognitive classification or performance in any of the five assessed cognitive domains, compared with the non-hallucinators. The Parkinson's disease groups also did not differ significantly in age, motor severity, medication use or duration of disease. On imaging analyses, the hallucinators, all of whom experienced visual hallucinations, exhibited grey matter atrophy with significant voxel-wise differences in the cuneus, lingual and fusiform gyri, middle occipital lobe, inferior parietal lobule, and also cingulate, paracentral, and precentral gyri, compared with the non-hallucinators. Grey matter atrophy in the hallucinators occurred predominantly in brain regions responsible for processing visuoperceptual information including the ventral 'what' and dorsal 'where' pathways, which are important in object and facial recognition and identification of spatial locations of objects, respectively. Furthermore, the structural brain changes seen on magnetic resonance imaging occurred independently of cognitive function and age. Our findings suggest that when hallucinators and non-hallucinators are similar in their cognitive performance, the neural networks involving visuoperceptual pathways, rather than the mesial temporal lobe regions, distinctively contribute to the pathophysiology of visual hallucinations and may explain their predominantly visual nature in Parkinson's disease. Identification of distinct structural MRI differences associated with hallucinations in Parkinson's disease may permit earlier detection of at-risk patients and ultimately, development of therapies specifically targeting hallucinations and visuoperceptive functions.