Abstract: Clinical data show evidence that a change in the arteriolar-to-venular diameter ratio (AVR) is associated with certain diseases such as cerebral atrophy, stroke, hypertension, and diabetes. An effective AVR measurement can be used to detect retinal diseases and pre-screen for their early diagnosis. This study presents a fully automated approach for measuring the AVR, which includes preprocessing the retinal image, automatically segmenting the vasculature, detecting the optic disc, determining the region of interest (ROI), classifying detected vessels into arteries and veins, measuring the vessel width, and calculating the AVR in the ROI. The method presented herein combines color features with structural features of vessel networks and proposes a novel feature vector based on the central line pixel of vessels, which leads to an improved classification of arteries and veins. Moeover, we propose a new region separation scheme for decreasing estimation errors in estimated AVR values, which uses 0.1 times the diameter of the optic disc as a step to separate the ROI. The detected vessels are then divided into five parts, and the estimated AVR value within these parts is averaged to obtain the final AVR estimation value. The proposed approach is evaluated against two different retinal image databases such as the publicly available DRIVE database and He Shi database, whose images acquired as part of the diabetic retinopathy screening program of He Shi ophthalmology hospital demonstrate better performance on vessel segmentation, artery/vein classification, and so on. In addition, when the evaluation value acquired by the proposed approach is compared with a reference AVR, a mean error of 0.04 was obtained in 20 test images provided in the He Shi dataset. It is evident that the estimated AVR values obtained using this approach gives results with a better agreement than that of recently developed methods.