Smooth pursuit eye movements are known to be driven by a mixture of visual feedback and predictive strategies. Prediction in pursuit allows humans to track predictable stimuli with minimal phase lag. But in certain disease states and focal neurological lesions, normal phase relationships are lost and humans track with increased phase errors. Using a sinusoidal pursuit paradigm, we sorted patients into those with large phase errors and those without. Then, working on the premise that large phase errors may have resulted from lack of prediction, we compared predictive and non-predictive ocular pursuit in patients with large phase errors, patients with normal phase errors and control subjects. Subjects sat in darkness and pursued an intermittently illuminated target moving with constant velocity to the right or left. When the movements were in alternate directions and predictable, all the groups possessed the ability to preprogramme appropriate anticipatory eye movements before target onset, and to use this for predictive pursuit. The difference between patients with large phase errors and normal subjects was not an absolute lack or possession of predictive ability but a difference in the timing at which a preprogrammed motor behaviour was initiated or terminated. The timing variability was wide and formed a graded continuum, the control subjects initiating anticipatory pursuit earlier, and the patients with large phase errors initiating much later. In a second experiment, subjects pursued a predictable ramp stimulus presented at various fixed frequencies. We found that in patients where anticipatory pursuit seemed abolished at one frequency of target presentation, changing the frequency of presentation elicited an anticipatory response. Patients adjusted their pursuit latencies to match the temporal demands of target presentation. At target frequencies above 0.8 Hz, there was a significant positive correlation between latencies in ramp pursuit and phase lags in sinusoidal pursuit. None of our patients showed complete loss of prediction irrespective of how large the phase errors were. Even when severe time delays in the system made it impossible for a subject to initiate anticipatory pursuit before target onset, prediction could still be demonstrated by the significant velocity and timing advantage the subject had in the pursuit of a predictable target stimulus or by the technique of unexpectedly blanking the target.