Abstract Background To our knowledge, no previous study assessed the reproducibility of non-mydriatic imaging of retinal microvessels using state-of-the-art Bland and Altman statistics. Methods In 194 subjects randomly selected from a Flemish population, we post-processed retinal images (Canon Cr-DGi) using IVAN software to generate the retinal arteriole and venule equivalents (CRAE and CRVE) and the arteriole-to-venule ratio (AVR). We searched for significant ( p ≤ 0.05) correlates of the retinal phenotypes, using stepwise multiple regression. To study intra- and inter-observer variability, 2 observers post-processed a random subset ( n = 84) in duplicate. According to Bland and Altman’s approach, reproducibility was twice the standard deviation of the pairwise differences between duplicate measurements, expressed as a percentage of the average of all measurements. Results In 194 participants (age range, 18–79 years; 52.5% women), CRAE, CRVE, and AVR averaged (±SD) 175.3 ± 22.8 μm, 284.3 ± 20.5 μm, and 0.62 ± 0.10, respectively. CRAE increased with female sex (+8.4 μm) and current smoking (+9.1 μm) and decreased with age (−0.30 μm/year) and mean arterial pressure (−0.41 μm/mmHg). CRVE decreased with age (−0.42 μm/year). AVR increased with female sex (+0.03 units). The intra-observer reproducibility of CRAE, CRVE and AVR was 13.2%, 8.4% and 9.0% for observer 1 and 10.3%, 10.8% and 16.0% for observer 2. Inter-observer reproducibility amounted to 10.8%, 9.9% and 14.6%, respectively. Conclusion Our study further validates the non-mydriatic approach to phenotype retinal microvessels, because it showed acceptable intra- and inter-observer variability and because the correlates of these retinal phenotypes were consistent with those reported in the literature.