Reported here is the agreement between three examination methods chosen to detect and grade diabetic retinopathy in 124 subjects with type II (noninsulin-dependent) diabetes mellitus. These three examination methods include ophthalmoscopy (indirect and direct) by a retina specialist, seven standard field fundus photographs read by the same retina specialist, and the same photographs read by a trained photographic grader at the Fundus Photograph Reading Center. For the 59 subjects examined with all three methods, these results indicated fair to good (kappas, 0.69-0.84) agreement between the retina specialist's and trained grader's reading of photographs, fair to good (kappas, 0.58-0.79) agreement between the retina specialist's ophthalmoscopic findings and the specialist's reading of photographs, and fair (kappas, 0.49-0.62) agreement between the retina specialist's ophthalmoscopic findings and the trained grader's reading of fundus photographs. Analysis of the disagreements confirmed earlier reports that ophthalmoscopy misses approximately 50% of eyes with microaneurysms only. Other disagreements resulted from the trained grader's overreading photographs of eyes with lesions simulating diabetic retinopathy. Of the 393 total subjects (diabetic and nondiabetic) in this study, such lesions were seen with ophthalmoscopy in six eyes of six subjects (2.4% of diabetic patients and 1.1% of nondiabetic subjects). The authors believe at least one definite retinal microaneurysm should be present in one eye before establishing the diagnosis of diabetic retinopathy in diabetic patients.