The motor manifestations of Parkinson's disease (PD) have been linked to an abnormal spatial covariance pattern involving basal ganglia thalamocortical pathways. By contrast, little is known about the functional networks that underlie cognitive dysfunction in this disorder. To identify such patterns, we studied 15 non-demented PD patients using FDG PET and a voxel-based network modeling approach. We detected a significant covariance pattern that correlated (p<0.01) with performance on tests of memory and executive functioning. This PD-related cognitive pattern (PDCP) was characterized by metabolic reductions in frontal and parietal association areas and relative increases in the cerebellar vermis and dentate nuclei. To validate this pattern, we analyzed data from 32 subsequent PD patients of similar age, disease duration and severity. Prospective measurements of PDCP activity predicted memory performance (p<0.005), visuospatial function (p<0.01), and perceptual motor speed (p<0.005) in this validation sample. PDCP scores additionally exhibited an excellent degree of test-retest reliability (intraclass correlation coefficient, ICC=0.89) in patients undergoing repeat FDG PET at an 8-week interval. Unlike the PD-related motor pattern, PDCP expression was not significantly altered by antiparkinsonian treatment with either intravenous levodopa or deep brain stimulation (DBS). These findings substantiate the PDCP as a reproducible imaging marker of cognitive function in PD. Because PDCP expression is not altered by routine antiparkinsonian treatment, this measure of network activity may prove useful in clinical trials targeting the progression of non-motor manifestations of this disorder.