The relative cost and cost-effectiveness of different methods of screening diabetic patients for sight-threatening retinopathy are assessed. The resource costs per screening visit, both to the health service and to patients, of ophthalmoscopic examination by primary screeners including general practitioners, hospital physicians, and ophthalmic opticians are estimated together with those of a similar screening test by ophthalmological clinical assistants. The total resource cost per screen of screening using non-mydriatic photography is also estimated. Using estimates of sensitivity, specificity, and prevalence generated in the screening of 3318 diabetic patients in three UK centres, the relative cost-effectiveness of screening methods is estimated in terms of their cost per true positive case detected. On the assumption that a patient makes a special trip for eye screening, the cost per true positive case detected for primary screeners ranges from 633 pounds for a GP-screened group in one centre to 1079 pounds for another GP-screened group in a second centre; the cost per true positive case detected of photography ranges from 497 pounds for a camera that is taken to general practices in one centre to 1546 pounds for a hospital-based camera. Relative cost-effectiveness changes if, in some contexts, the screening can take place without requiring an additional patient visit, and is strongly related to the relative sensitivity of the screening methods and to the prior probability (prevalence or incidence) of retinopathy in the diabetic population.