The performance of 19 general practitioners in a written simulation of patient-doctor encounters was compared with real data on prescription and referral from 'sick-fund' patients. Two issues were studied: the amounts of certain drugs prescribed (analgesics and antirheumatics; hypnotics/sedatives and tranquillizers; neuroleptics and antidepressants) and the number of referrals to specialists. In the simulation, experts assessed the 'risk of unnecessary harm to the patient' induced by the therapeutic procedures of these general practitioners, as described previously. A close correlation was found between the ratings of the risk of unnecessary harm in the simulation and the prescription and referral data in reality for general practitioners who were consistently generous or frugal in prescribing the above drugs and referring the patients to specialists. The tentative conclusion is that the written simulation with its rating procedure discriminates reliably between general practitioners who induce less risk of unnecessary harm from those who are likely to cause harm to the patient.