Diabetic Retinopathy (DR) is one of the leading causes of preventable blindness in the developed world. As the number of diabetic patients grows, there is an increasing need for automatic DR detection screening systems. We propose to segment DR lesions using Conditional Generative Adversarial Networks where an adversarial loss is added on top of the segmentation loss. We show that the addition of the adversarial loss improves the lesion segmentation performance of the resulting models. For instance, our approach improves the results of a Holistically-Nested Edge Detection (HEDNet) model on the task of Hard Exudates’ segmentation from 79.35% Average Precision (AP) to 81.83% AP. We show improvements on microaneurysms, hemorrhages and hard exudates segmentation, which are the most important lesions to perform DR detection and grading.