Several studies have concluded that specialists form a hidden system for primary-care delivery. However, these studies assume that a specialist who provides the majority of care is the primary-care physician. Using data for a one-year period from 2752 people enrolled in the Rand Health Insurance Experiment, we examined the validity of this conclusion. We compared the effects of three different definitions of a primary-care physician on identification of the primary-care provider: the physician who delivered the "majority of care" (34 per cent were specialists), the physician designated by the patient to receive the results of a multiphasic-screening examination (12 per cent were specialists), and the physician who treated common problems (9 per cent were specialists). Use of the "majority-of-care" criterion to define primary care overestimated by threefold the contribution specialists make to this activity. Definitions of a primary-care physician must be more specific and should include the tasks frequently associated with primary care, as well as patients' perceptions of the physician who provides their primary care.