Introduction We must study alternatives to structure an effective diabetic retinopathy screening program for Brazilian public health system. Objectives Evaluate the diagnostic performance of retinal digital photography for diabetic retinopathy screening in primary care, accuracy of the family physician in diabetic retinopathy identification compared to the ophthalmologist, and the need for dilation. Methodology In a primary care service were performed retinal photographs with non-mydriatic Retinal Camera in 219 type 2 diabetic patients with and without medication mydriasis. We evaluated the performance of the diagnostic of the photos graded by three family physicians with training compared to two ophthalmologists (gold standard), and explore related factors with the need for mydriasis pharmacologically. Results The prevalence of diabetic retinopathy and proliferative diabetic retinopathy was 19.2% and 1.5%, respectively. The sensitivity of family physicians to evaluate diabetic retinopathy averaged 82.9% (66.7–94.8%); specificity, 92% (90.2–93.3%); the accuracy, 90.3% (88.2–93%) and positive predictive value, 71.2% (68–75.5%). The agreement calculated using the kappa adjusted coefficient was from 0.74 to 0.8 for retinopathy and 0.88 to 0.92 for macular edema. Without drug mydriasis the photos were unreadable by 14.8%, when using mydriatic collyrium this number decreased to 8.7% (McNemar test, P < 0.005). Patients with more than 65 years old has more readability after drug mydriasis (McNemar test, P = 0.011). Conclusion Trained family physician reached a good performance for evaluation of retinography for diabetic retinopathy. There was improvement in readability with pupil dilation in older patients.