The Movement Disorder Society Task Force (MDS-TF) has proposed diagnostic criteria for mild cognitive impairment in Parkinson's disease (PD-MCI). We hypothesized that the risk of dementia (PDD) varies across the different cutoff schemes allowed. A longitudinal study followed 121 non-demented PD patients for up to 4.5 years. In Part One, unique groups of patients were identified as PD-MCI at baseline using the MDS-TF requirement of two impaired cognitive test scores, with both scores classified as impaired at either (i) 2 s.d., (ii) 1.5 s.d. or (iii) 1 s.d. below normative data; relative risk (RR) of PDD was assessed at each criterion. In Part Two, the whole sample was reassessed and (i) RR of PDD determined when two impairments at 1.5 s.d. existed within a single cognitive domain, followed by (ii) RR of PDD in the unique group whose two impairments at 1.5 s.d. did not exist within a single domain (i.e., only across two domains). Twenty-one percent of patients converted to PDD. Part One showed that the 1.5 s.d. criterion at baseline is optimal to maximize progression to PDD over 4 years. Part Two, however, showed that the 1.5 s.d. cutoff produced a high RR of PDD only when two impairments were identified within a single cognitive domain (7.2, 95% confidence interval (CI)=3.4-16.6, P<0.0001; 51% converted). The RR when the 1.5 s.d. impairments occurred only across two different domains, was nonsignificant (1.7, CI=0.5-7.4, P=0.13; 11% converted) and similar to using a 1 s.d. criterion (1.9, CI=0.3-4.3, P=0.13; 8% converted). If the intent of a PD-MCI diagnosis is to detect increased risk of PDD in the next 4 years, optimal criteria should identify at least two impairments at 1.5 s.d. within a single cognitive domain.