Abstract Aim Retinal vessel diameters are candidate biomarkers of mortality prediction in large population-based studies. We aimed to investigate the predictive value of retinal vessel diameters and flicker-induced retinal arteriolar and venular dilation on all-cause mortality in long-term follow-up of haemodialysis patients. Methods and results Retinal vessel diameters as well as maximum arteriolar (aMax) and venular dilation (vMax) were investigated in 275 and 214 haemodialysis patients, respectively. Patients were observed in a long-term follow-up for a median period of 73 months. About 36% (76/214) and 41% (113/275) of patients died. Arteriolar and venular diameters were 175 ± 19 and 208 ± 20 µm, respectively. Median aMax and vMax were 1.6 (0.3–3.3) and 3.2 (2.0–5.1)%. Patients within the lowest tertile of vMax showed lower 5-year survival rates compared with the highest tertile (50.6 vs. 82.1%) and also exhibited a higher incidence of infection-related deaths (21.7 vs. 4.0%). Univariate hazard ratio (HR) per standard deviation increase of vMax for all-cause mortality was 0.69 (0.54–0.88) and was even more pronounced for infection-related mortality [HR 0.53 (0.33–0.83)]. Regarding all-cause mortality, multivariate adjustment for eight non-retinal mortality predictors including interleukin-6 did not attenuate the HR relevantly [0.73 (0.54–0.98)]. Arteriolar and venular diameters did not predict all-cause nor cardiovascular and infection-related mortality. Conclusions Long-term follow-up of patients on haemodialysis demonstrated the potential of retinal venular dilation capacity for mortality prediction, which was most pronounced for infection-related mortality. In the same cohort, retinal arteriolar and venular diameters showed no predictive value for hard endpoints. Retinal venular dilation but not arteriolar and venular diameters is a valuable diagnostic biomarker for risk prediction in patients with end-stage renal disease and should be considered for monitoring of critically ill patients.