For Parkinson's disease (PD), little is known about how the choice of diagnostic criteria affects research results. Using data on PD from three community studies (from Argentina, the Netherlands, Italy), we compared the impact on prevalence of several sets of diagnostic criteria. Each set was based on cardinal signs--resting tremor, bradykinesia, rigidity, impaired postural reflexes--and required that other parkinsonism be excluded. Some sets had additional requirements related to duration of symptoms, asymmetry of signs, or response to medication. In terms of prevalence, much lower estimates were associated with the requirements of asymmetry of signs and response to medication. The assessment of these clinical features may not be practical in community studies. Impaired postural reflexes, as a cardinal sign, seemed superfluous. For community studies of PD, we recommend the following diagnostic criteria: at least two of resting tremor, bradykinesia, or rigidity, in the absence of other apparent causes of parkinsonism.