As part of a genetic study, we investigated the accuracy of a telephone questionnaire to diagnose or screen for Parkinson's disease (PD). We studied 20 PD patients randomly selected among all incident cases in Olmsted County, Minnesota, for the period 1970 through 1988. Each patient was matched by age and sex to a subject free of PD or parkinsonism residing in the same county. In addition, we studied 20 patients with PD referred to our institution from outside of Olmsted County between 1991 and 1993. Medical record documentation of disease status for both affected and unaffected subjects served as the standard for comparison. A trained interviewer kept unaware of disease status administered nine symptom questions via telephone to all study subjects (or their proxy). Seventeen PD cases from Olmsted County, all 20 referral cases, and 16 unaffected subjects participated in the interview (total = 37 with PD and 16 unaffected subjects); 36 interviews (24 PD) were with a proxy and 17 (13 PD) were direct. Questions regarding "poor balance," "trouble buttoning," and "trouble arising" had the highest sensitivity; questions regarding "shaking," "feet stuck to the floor," "softer voice," and "smaller writing" had the highest specificity. A combination of any four symptoms yielded the best balance between sensitivity (89%) and specificity (88%). Although the questionnaire appears to be useful both as a screening and as a diagnostic tool, it failed to reach complete accuracy, and it should be used in the context of a more complex case-finding strategy.