Patients with Parkinson's disease (PD) show impairments on tasks that require them to switch attention between two perceptual dimensions (extradimensional (ED) shifting). It has been suggested that ED shifting deficits can be caused by two separate mechanisms, 'learned irrelevance' and 'perseveration'. This study set out to test the hypothesis that enhanced learned irrelevance is present in medicated patients with PD. An enhancement of learned irrelevance in PD patients should result in increased errors on a 'deficit' shift relative to controls and decreased errors on an 'improvement' shift. A similar pair of deficit and improvement shifts were used to detect possible enhanced perseveration in patients. Instead of showing the predicted patterns of deficit and improvement, patients displayed a consistent deficit on those shifts that required that they switch their attention to a different dimension (ED shifts). In contrast, patients were not impaired on shifts that required no such shift of attention (intradimensional shifts). Although there was an increase in errors at the learned irrelevance deficit shift, a similar increase at the learned irrelevance improvement shift shows that enhanced learned irrelevance is not responsible for either of these results. Patients were no more distractible than controls, but displayed increased 'loss of set' as measured by errors generated after a rule was learned. These results point to the existence of exaggerated, rigid selective attention in patients with PD rather than a breakdown in the ability to selectively attend. There was no evidence for the existence of enhanced learned irrelevance in the patients.