Deep brain stimulation of the subthalamic nucleus improves motor functions in patients suffering from advanced Parkinson's disease but in some patients, it is also associated with a mild decline in cognitive functioning about one standard deviation from the preoperative state. We assessed the impact of the cortical lead entry point, the subcortical electrode path and the position of the active electrode contacts on neuropsychological changes after subthalamic nucleus-deep brain stimulation compared to a control group of patients receiving best medical treatment. Sixty-eight patients with advanced Parkinson's disease were randomly assigned to have subthalamic nucleus-deep brain stimulation or best medical treatment for Parkinson's disease. All patients had a blinded standardized neuropsychological exam (Mattis Dementia Rating scale, backward digit span, verbal fluency and Stroop task performance) at baseline and after 6 months of treatment. Patients with subthalamic nucleus-deep brain stimulation were defined as impaired according to a mild decline of one or more standard deviations compared to patients in the best medical treatment group. The cortical entry point of the electrodes, the electrode trajectories and the position of the active electrode contact were transferred into a normalized brain volume by an automated, non-linear registration algorithm to allow accurate statistical group analysis using pre- and postoperative magnetic resonance imaging data. Data of 31 patients of the subthalamic nucleus-deep brain stimulation group and 31 patients of the best medical treatment group were analysed. The subthalamic nucleus-deep brain stimulation group showed impaired semantic fluency compared with the best medical treatment group 6 months after surgery (P = 0.02). Electrode trajectories intersecting with caudate nuclei increased the risk of a decline in global cognition and working memory performance. Statistically, for every 0.1 ml overlap with a caudate nucleus, the odds for a decline >1 standard deviation increased by a factor of 37.4 (odds ratio, confidence interval 2.1-371.8) for the Mattis Dementia Rating Scale and by a factor of 8.8 (odds ratio, confidence interval 1.0-70.9) for the backward digit span task. Patients with subthalamic nucleus-deep brain stimulation who declined in semantic verbal fluency, Stroop task and the backward digit span task performance showed a position of the active electrode outside the volume built by the active electrodes of stable performers. Passage of the chronic stimulation lead through the head of the caudate increases the risk of global cognitive decline and working memory performance after subthalamic nucleus-deep brain stimulation in Parkinson's disease. Therefore the electrode path should be planned outside the caudate nuclei, whenever possible. This study also stresses the importance of precise positioning of the active stimulating contact within the subthalamic volume to avoid adverse effects on semantic verbal fluency and response inhibition.