Existing studies on task switching in Parkinson's disease (PD) patients have led to somewhat different results. In particular, it is unclear whether PD patients have a deficit in attentional control. In this study, we assessed task-switching abilities in samples of non-demented PD patients and elderly controls. We used a paradigm in which there was a random task sequence and the task was cued in every trial. This allowed the investigation of both task-set reconfiguration and task-set dissipation. In terms of the proportion of errors made, the patients showed increased switch cost and congruency effects. For reaction times, PD patients showed enlarged congruency effects on switch trials, specifically in the condition in which we used a short constant response-cue interval (RCI). Nevertheless, in a similar fashion to older controls, the patients showed reductions in reaction time switch cost from a short to a long cue-target interval (CTI) and from a short to a long RCI. While these latter findings, respectively, suggest unimpaired task preparation and task dissipation on correct trials in the PD patients, the overall results show that they have a deficit in biasing and selecting currently relevant task sets and more generally argue in favour of a failure of attentional control in PD.