Screening asymptomatic subjects for diabetes is often undertaken using a random capillary whole blood sample for glucose estimation. The test characteristics of this method for screening were assessed using a glucose oxidase method among 3425 Europeans and 3469 South Asians who gave such a sample during the Coventry Diabetes Study, a house-to-house diabetes prevalence study. Glucose tolerance test were performed on those with a high blood glucose and 10% of others. Previously undiagnosed diabetes was found in 73 Europeans and 110 South Asians. If the random glucose was > or = 7.0 mmol l-1, 8.0% of Europeans and 6.7% of South Asians would need a further diagnostic test and the sensitivity of this cut-off was 51.7 (95.0% CI: 43.5-59.9)% in Europeans and 68.4 (60.6-76.2)% in South Asians. Sensitivity was increased in South Asians but not Europeans by defining the time since last meal (South Asians < 2 h: 83.9 (72.3-92.0)%; > or = 2 h: 54.9 (42.7-66.8)%). Sensitivity was poorest among Europeans aged > or = 65 years (40-64 years 69.0 (49.2-84.7)%, > or = 65 years 49.4 (38.2-60.6)%). Screening asymptomatic individuals using an isolated capillary random whole blood glucose measure is a poor test for diabetes, although slight improvement can be obtained among South Asians by testing within 2 h of a meal.