The purpose of this study was to investigate the neuropsychological correlates of pathological gambling (PG) in Parkinson's disease (PD). Fifteen patients with PD affected by PG (identified based on DSM-IV criteria; PD+PG) without clinically evident dementia were compared with 15 nondemented patients with PD not affected by PG (PD-PG). Two groups of patients with PD were matched for age, length of education, and gender. Clinical and neuropsychiatric features were assessed; several cognitive domains, mainly related to executive functions, were explored by means of standardized neuropsychological tasks. PD+PG and PD-PG did not differ on clinical and neuropsychiatric aspects. PD+PG patients performed significantly worse than PD-PG patients on cognitive tasks that evaluated visuo-spatial long-term memory and several frontal lobe functions. After Bonferroni correction, differences remained significant on the Frontal Assessment Battery (FAB) (P = 0.001), on phonological fluency task (P = 0.003), and on the Trail Making Test, part B minus part A (P = 0.002). Logistic regression analysis demonstrated that low scores on the FAB were the only independent predictor of PG (odds ratio, 27.9; 95% CI: 2.82-277.95, P = 0.004). The results indicate an association between PG and frontal lobe dysfunctions in nondemented patients with PD. Low scores on the FAB indicate patients with PD at high risk for PG.