Eye movement desensitization is a new and effective procedure for post-traumatic stress disorder that requires explanation. Reverse learning is a model developed in artificial neural networks as a theoretical explanation of rapid-eye-movement sleep. It demonstrates that an overloaded node within a network can be consolidated with a series of non-specific activations. Rapid-eye-movement sleep is suspected to have a memory consolidation function. Ponto-geniculo-occipital spikes, which occur in rapid-eye-movement sleep, are a candidate for such activations in the real brain. In cats, the phasic functions of rapid-eye-movement sleep are driven by ponto-geniculo-occipital spikes, which are non-specific, at highest amplitude in the visual system but present in other parts of the cortex. Such spikes can be evoked by sensory events such as eye movements. There is evidence of similar events in the human brain. Induced eye movements could generate ponto-geniculo-occipital equivalent spikes and eye movement desensitization/reprocessing could be explained as a focused and artificial exploitation of the rapid-eye-movement sleep mechanism. This theory of eye movement desensitization/reprocessing enables some explanation of current results and may be relevant to other problems, such as stereotyped behaviour.