Background: Impulse control disorders (ICDs) and related behaviors are frequent in Parkinson's disease (PD). Mild cognitive impairment (PD-MCI) and dementia (PDD), both characterized by heterogeneous cognitive phenotypes, are also commonly reported in PD. However, the frequency and severity of ICD within PD cognitive states is unknown. Methods: Three hundred and twenty-six PD patients completed a comprehensive neuropsychological assessment and were classified as PD-MCI, PDD, or without cognitive alterations (PD-NC). The Minnesota impulsive disorders interview was used to ascertain the presence (ICD+) or absence (ICD-) of ICD. The Questionnaire for Impulsive-Compulsive Disorders in Parkinson's Disease-Rating Scale was used to assess ICD severity. A subsample of 286 patients evaluated with the same cognitive tasks was selected in order to investigate the characteristics of ICD in PD cognitive phenotypes. Results: ICDs were present in 55% of PD-NC, in 50% of PD-MCI, and in 42% of PDD patients. Frequencies of ICD+ with attentive (ICD+: 20% vs. ICD-: 4%; p = 0.031) and executive impairments (ICD+: 44% vs. ICD-: 30%; p = 0.027) were higher in the PD-MCI and PDD subgroups, respectively. As expected, no differences were observed in the PD-NC. PD-MCI with attentive impairments presented higher percentage of ICD+ with deficits in the Trail Making Test B-A but not in the Digit Span Sequencing task. In PDD, executive failures concerned Similarities task (ICD+: 67%; ICD-: 29%; p = 0.035), with no differences between ICD+ and ICD- in the Stroop task. Conclusions: Prevalence and severity of ICDs and related behaviors do not differ in PD with different cognitive states. However, ICD+ are more likely to show deficits, respectively in attentive and in executive domains, specifically in the Trail Making Test B-A task for the attention and working memory domain in PD-MCI and in the Similarities task for the executive function domain in PDD. Prospective studies should evaluate if these tests can be used as screening tool for ICDs in PD.