Following the thalidomide tragedy, schemes designed to monitor the safety of drugs were established in several countries. However, in the late 1970s, renewed concern about the effectiveness of the current post-marketing surveillance (PMS) methods led to new schemes being proposed. Although none of these was implemented in this country, they did lead to the development of Prescription-Event Monitoring (PEM) within the University of Southampton. PEM is the first large scale PMS method used in general practice to link drug exposures (identified by means of prescriptions) with subsequent events recorded in patients' case notes. The objectives of this research were to assess (a) the extent of the cooperation of general practitioners with PEM, (b) the feasibility of event reporting and whether it creates any problems with patient confidentiality, (c) whether it is possible to measure relative rates of events experienced by patients treated with various drugs, (d) whether PEM will be able to detect adverse reactions earlier than is currently possible with voluntary reporting systems, and (e) the cost of PEM. The methodology and a detailed analysis of the results of the PEM pilot study, which compared benoxaprofen and fenbufen, two non-steroidal anti-inflamatory drugs (NSAIDs), are presented. The digestive disorders recorded in the pilot study are compared with those recorded in a second study of benoxaprofen and fenbufen and in subsequent studies of three other NSAIDs (piroxicam and two formulations of indomethacin). The results of a hypothesis-testing PEM study are also discussed. The response of general practitioners to the questionnaires requesting details of events experienced by patients treated with the study drugs ranged from 51% in the pilot study to 76% in the hypothesis-testing study; this was considered to be very satisfactory. Almost all the responding doctors understood the concept of event reporting, which appeared to create no confidentiality problems. It was possible to compare relative rates of events experienced by patients treated with NSAIDs; the known adverse effects of the drugs (as well as a previously unknown effect of benoxaprofen) stood out clearly against a background of unrelated events. However, the analysis was not without complications because the dates of the events were not recorded on the computer. PEM is subject to a minimum delay of three months from the time that a drug is prescribed to the time of mailing the questionnaire. Therefore, voluntary reporting systems might, in theory, detect an adverse reaction before PEM does. However, the present poor cooperation with the former makes this unlikely. PEM is relatively inexpensive, especially when compared with the cost of PMS studies organised by pharmaceutical companies. It is concluded that PEM is an important advance in the study of drug safety.