The results of the screening of 3318 diabetic patients for sight-threatening diabetic retinopathy in three UK centres are reported. The aims of the study were to determine the extent of diabetic retinopathy in the screened population and to assess the relative effectiveness of different screening methods in appropriately referring cases from a diabetic population, in a context very close to a routine clinical service. Patients were assessed by ophthalmoscopic examination by an ophthalmological clinical assistant. The clinical assistants' referral grades formed the reference standard against which to assess the effectiveness of other screening methods including ophthalmoscopy by primary screeners who were general practitioners (GPs), ophthalmic opticians and hospital physicians, and the assessment by consultant ophthalmologists of non-mydriatic Polaroid fundus photography. The performance of primary screeners based on ophthalmoscopy ranged from a sensitivity of 0.41, with a specificity of 0.89, for one of the GP groups, to a sensitivity of 0.67, with a specificity of 0.96, for the hospital physician group. The performance of the non-mydriatic camera ranged from a sensitivity of 0.35, with a specificity of 0.95, to a sensitivity of 0.67, with a specificity of 0.98.