Generalisability theory is presented as a research method for assessing the quality of health advice. The theory is applied to make a comprehensive assessment of primary health care advice given by community pharmacists. A random sample of pharmacies from all London postal districts were selected and visited for set periods throughout one year. During these visits all consultations on health or the use of drugs between pharmacists and clients were tape-recorded. Nine characteristics were derived by an external criterion panel on which the quality of advice in a random sample of consultations (50) was assessed using rating scales. Following the application of generalisability theory, operational conditions for the assessment were selected to obtain a generalisability coefficient of around 0.8. Nearly half the consultations achieved satisfactory scores on at least three-quarters of the criteria on which they were assessed. However almost a third were considered satisfactory on less than a quarter of the criteria, most of these being satisfactory on none of them. Performance regarding disease prevention/health promotion activity was poorest. Thus the quality of most consultations was found to be either very good or very poor, relatively few occupying the middle ground.