To investigate the pathophysiology of dystonia, we recorded contingent negative variation (CNV) in 12 patients with cervical dystonia and in 12 age-matched normal subjects. In a simple reaction time paradigm, the subjects were given a pair of a warning stimulus and a subsequent stimulus that triggered head rotation to either side or extension of the fingers. In normal subjects, CNVs for head rotation were not affected by neck muscle pre-activation simulating torticollis, and were always symmetrical with equally high amplitudes over the frontal and central leads. By contrast, CNVs for finger movement had the maximum over the central lead and showed a characteristic distribution; those for the right finger movement had the left hemisphere dominance, whereas those for the left finger movement had similar amplitudes over both hemispheres. In patients with rotatory torticollis (rotatocollis), the components of CNVs for head rotation were markedly attenuated in all the leads, except for the initial negative deflection (orienting response). As a whole, cervical dystonia patients had significantly lower amplitudes of late CNVs for head rotation than normal subjects (P < 0.001), whereas late CNV amplitudes in finger extension did not differ in the two groups. Their reaction times for head rotation were similar, but durations of EMG activities were prolonged in the patients because of co-contractions of the antagonists. The task-specific CNV amplitude loss is therefore not explained by reaction times or by the abnormal neck muscle activities prior to the movement, but it reflects a failure of neural activities preparing for a phasic neck movement, resulting in co-contraction of the agonists and the antagonist. Dystonia may be associated with defective retrieval or retaining of specific motor programmes or subroutines in response to sensory stimuli.