- Abstract: Today, the analysis of retina images is a complex, manual task requiring a highly-skilled clinician. In light of the recent successes of Fully Convolutional Networks (FCNs) applied to biomedical image segmentation, we want to assess its potential in the context of retinal artery-vein (A/V) discrimination. With the aim of improving the automation of vessel analysis, a novel application of the U-Net semantic segmentation architecture (based on FCNs) on the discrimination of arteries and veins in fundus images is presented. By utilizing deep learning, results are obtained that exceed accuracies reported in the literature. Our model was tested on the public DRIVE dataset, measuring performance on vessels wider than two pixels, achieving accuracies of 94.42% and 94.11% on arteries and veins, respectively. This represents a decrease in error of 25% over the previous state of the art reported by Xu et al. (2017). Fully Convolutional Networks combined with careful data augmentation do foster potential in A/V discrimination on a small data set, outperforming previous work. Evaluation masks and predicted A/V annotations on the public DRIVE data set are available at http://iflexis.com/DRIVEmasks.