In recent years, several attempts have been made to characterize the nature of the cognitive deficits shown by patients with Parkinson's disease. It has been suggested variously that they have difficulty in switching cognitive set, in performing effortful (or controlled) as opposed to automatic tasks, or that their impairment is found in tasks which maximize the amount of 'self-directed task specific planning'. It is proposed that this latter distinction may be reformulated in terms of the degree of internal versus external attentional control which is required by the task. An experiment is described which attempted to manipulate this parameter. A version of the Stroop colour-word test was used, in which the words 'red' and 'green' were presented in the complementary coloured 'ink'. Subjects responded either to the colour of the ink in which the word was written or the colour named by the word. The relevant attribute changed at intervals during the course of the experiment. In one condition, the relevant stimulus attribute was cued before each trial. In another condition, subjects had to remember which attribute was currently relevant. Results revealed that patients with Parkinson's disease were impaired mainly on the second version of the task which required internal attentional control. The results are discussed in relation to the models of Working Memory (Baddeley, 1986), and attentional control (Norman and Shallice, 1980). Exploration of these models leads to the formulation of a theory in which the crucial determinant of cognitive impairment in Parkinson's disease is reduced resources in the Supervisory Attentional System. Provided the demands of the task are within the patient's available attentional resources the patient may not show any deficit. If, however, the attentional demands exceed available resources, as in tasks which depend upon internal cues, then deficits will be observed.