The proportion of attenders at accident and emergency (A&E) departments who present 'inappropriately' with conditions which could be managed in general practice has been estimated at between 6.7 per cent and 64-89 per cent. These estimates have been based on subjective assessments by clinicians, or on an objective classification developed by the Nuffield Provincial Hospitals Trust (NPHT). This study sought to validate this classification, and to develop and validate other objective systems of classifying A&E attenders. Two novel methods were devised, one based on the ICD-9 diagnosis and one primarily on processes of care. All three techniques were validated against the pooled opinions of a sample of general practitioners (GPs). The existing NPHT classification was found to be very unreliable. The new diagnostic method was more specific but had poor sensitivity, whereas the technique based on processes of care agreed remarkably well with the sample of GPs. This method was applied retrospectively to random samples of 8877 adult self-referrals to 16 English A&E departments, and yielded an estimate that 23 per cent could have been treated in general practice. This approach provides a simple and valid retrospective method for identifying patients who were suitable for care in general practice. The method may be used to identify groups of patients who frequently attend inappropriately, to identify areas in which primary care needs are not being effectively met in general practice, and to form a basis for planning and auditing strategies to meet those needs in a more appropriate setting.