As many as two million people in the United Kingdom repeatedly see people, animals, and objects that have no objective reality. Hallucinations on the border of sleep, dementing illnesses, delirium, eye disease, and schizophrenia account for 90% of these. The remainder have rarer disorders. We review existing models of recurrent complex visual hallucinations (RCVH) in the awake person, including cortical irritation, cortical hyperexcitability and cortical release, top-down activation, misperception, dream intrusion, and interactive models. We provide evidence that these can neither fully account for the phenomenology of RCVH, nor for variations in the frequency of RCVH in different disorders. We propose a novel Perception and Attention Deficit (PAD) model for RCVH. A combination of impaired attentional binding and poor sensory activation of a correct proto-object, in conjunction with a relatively intact scene representation, bias perception to allow the intrusion of a hallucinatory proto-object into a scene perception. Incorporation of this image into a context-specific hallucinatory scene representation accounts for repetitive hallucinations. We suggest that these impairments are underpinned by disturbances in a lateral frontal cortex–ventral visual stream system. We show how the frequency of RCVH in different diseases is related to the coexistence of attentional and visual perceptual impairments; how attentional and perceptual processes can account for their phenomenology; and that diseases and other states with high rates of RCVH have cholinergic dysfunction in both frontal cortex and the ventral visual stream. Several tests of the model are indicated, together with a number of treatment options that it generates.

The dual-deficit model of visual hallucinations (Collerton et al. target article) is compared with the dual-deficit model of auditory hallucinations (Waters et al., in press). Differences in cognitive mechanisms described may be superficial. Similarities between these models may provide the basis for a general model of complex hallucinations extended across disorders and modalities, involving shared (overlapping) cognitive processes.

A “multifactorial” model should accommodate a psychological perspective, aiming to relate the phenomenology of complex visual hallucinations not only to neurobiological findings but also an understanding of the patient's psychological problems and situation in life. Greater attention needs to be paid to the role of the “lack of insight” patients may have into their hallucinations and its relationship to cognitive impairment.

Recurrent complex visual hallucinations (RCVH) represent a form of psychosis. It may be useful to compare RCVH to another form of psychosis, catatonia. Both include a long list of medical illnesses and have been examined using several different hypotheses. Catatonia has a variety of hypotheses, including neurocircuitry, neurochemistry, and an integrated neuropsychiatric hypothesis. This hypothesis for catatonia supports Collerton et al.'s Perception and Attention Deficit model (PAD) for RCVH.

The Perception and Attention Deficit (PAD) model of visual hallucinations is as limited in generality as other models. It does, however, raise an interesting hypothesis on the role of attentional biases among proto-objects. The prediction that neither impaired attention nor impaired sensory activation alone will produce hallucinations should be addressed in future studies by analysing partial correlations between putative causes and hallucinatory effects.

Collerton et al.'s Perception and Attention Deficit (PAD) model argues that all recurrent complex visual hallucinations (RCVH) result from maladaptive, deficient sensory and attentional processing. We outline a constructivist-based representation of perception using signal detection theory, in which hallucinations are modeled as false alarms when confirmational perceptual information is lacking. This representation allows for some individuals to have RCVH due to a criterion shift associated with attentional proficiency that results in an increased awareness of the environment.

Dichotomizing perceptions, by those that have an objective reality and those that do not, is rejected. Perceptions are suggested to fall along a multidimensional continuum in which neither end is totally “pure.” At the extreme ends, perceptions neither have an objective reality without some subjectivity, nor, at the other end, even as hallucinations, are they totally dissociated from reality.

When viewed from a distance, visual hallucinations fall into one of two symptom patterns, a dichotomy which poses a problem for theoretical models treating them as a single entity. Such models should be broadened to allow for two distinct but overlapping syndromes – one likely to relate to visual de-afferentation, the other to Perception and Attention Deficit (PAD) cholinergic pathology.

This commentary takes a closer look at how “constructive models of subjective perception,” referred to by Collerton et al. (sect. 2), might contribute to the Perception and Attention Deficit (PAD) model. It focuses on the neuronal mechanisms that could mediate hallucinations, or false inference – in particular, the role of cholinergic systems in encoding uncertainty in the context of hierarchical Bayesian models of perceptual inference (Friston 2002b; Yu & Dayan 2002).

There are strong resemblances between the neurobiological characteristics of hallucinations occurring in the particular case of schizophrenia and the hallucinatory activity observed during the rapid-eye-movement (dreaming) sleep stage: the same prefrontal dorsolateral deactivation; forebrain disconnectivity and disinhibition; sensory deprivation; and acetylcholine, monoamine, and glutamate modifications.

There is little to refute in Collerton et al.'s argument that recurrent complex visual hallucinations involve multiple physiological mechanisms, and the target article's proposed PAD model implicitly incorporates this concept, advancing the field. The novel concept in this model is the intrusion of hallucinatory proto-objects into relatively preserved scenes. The weakness of the model is the lack of physiological detail for this mechanism.

Collerton et al. postulate that in a variety of different clinical conditions, hallucinations are derived from object schema lodged in long-term memory. I review two new experiments in which memory images can be easily triggered in neurologically intact subjects. These examples of making visible items in memory may provide experimental models for genesis of hallucinations.

The role of brain monoamines may be important for the neurobiology of the alterations of visual alertness in recurrent complex visual hallucinations (RCVH). This is evidenced by sleep research, neurophysiologic, and clinical data. Hence, the mechanisms of RCVH may not be simply explained by acetylcholine underactivity only.

Recent research on visual mental imagery plays an important role for the study of visual hallucinations. Not only are mental images involved in various cognitive processes, but they also share many processes with visual perception. However, we rarely confuse mental images with percepts, and recent neuroimaging studies shed light on the mechanisms that are differently activated in imagery and perception.

Before a general cognitive model for recurrent complex visual hallucinations (RCVH) is accepted, there must be more research into the neuropsychological and cognitive characteristics of the various disorders in which they occur. Currently available data are insufficient to distinguish whether the similar phenomenology of RCVH across different disorders is in fact produced by a single or by multiple cognitive mechanisms.

Waking hallucinations suggest mechanisms of dream initiation and maintenance. Visual association cortex activation, yielding poorly attended-to, visually ambiguous dream environments, suggests conditions favoring hallucinosis. Attentional and visual systems, coactivated during sleep, may generate imagery that is inserted into virtual environments. Internally consistent dreaming may evolve from successive, contextually evoked images. Fluctuating arousal and context-evoked imagery may help explain dream features.

Collerton et al. propose that one and the same mechanism (PAD) underlies recurrent complex visual hallucinations (RCVH) in various disorders, including schizophrenia, dementia, and eye disease. The present commentary offers an alternative account of RCVH and other recurrent complex hallucinations specific to schizophrenia and related disorders only. The proposed account is consistent with the bias of schizophrenic RCVH contents toward animate, socially active entities.

This commentary reviews and extends the target article's treatment of the topic of the role of acetylcholine in hallucinatory experience in health and disease. Particular attention is paid to differentiating muscarinic and nicotinic effects in modulating the use of virtual reality mechanisms by the brain. Then, attention is drawn to the similarities between these aspects of brain function and certain aspects of television digital compression technology.

We tested Collerton et al.'s model of visual hallucinations by re-examining a data set for correlations between visual hallucinations and measures of attentional function in schizophrenia patients. These data did not support their model. We suggest that cortical hyperexcitability plays an important role in hallucinations, and propose an alternative model that links evidence for cortical hyperexcitability with abnormal neural dynamics.

The etiology of visual hallucinations is largely undetermined in schizophrenia. Collerton et al.'s PAD model partly concurs with what we know about neurocognition in schizophrenia, but we need to specify the types of perceptual and attentional abnormalities that are implicated in recurrent complex visual hallucinations (RCVH). Available data suggest that abnormal attentional control and top-down processing play a larger role than the ventral stream deficits.