Because cognitive impairment might pose a safety risk for these drivers and the public, we examined what patient characteristics might be associated with motor vehicle crashes, in a retrospective study of all new referrals to a geriatric clinic specializing in memory and behavior problems between July 1, 1990 and June 30, 1995. During this time, 989 new patients were evaluated by the clinic staff. Driving data were available from 634 patients. The only factor significantly associated with crashes was driving alone (odds ratio = 2.23, 95% confidence interval = 1.20-4.15). Twenty-five percent of patients who drove alone had caregiver-reported crashes in the previous 5 years, compared with 13% of those who drove only with a passenger. Patients who drove alone were more likely to have spousal caregivers than patients who drove only with a passenger (72% vs. 58%, p = 0.028). Thus, cognitively impaired patients allowed to drive alone were more likely to have been involved in crashes than patients not driving alone. However, the causal nature of this association cannot be established with the present design. Prospective studies are required to determine if the presence of a co-pilot represents a safe strategy to extend driving privileges in cognitively impaired older drivers.