Summary Background Endothelial dysfunction is a marker of future cardiovascular disease (CVD) risk, yet epidemiological studies have yielded inconsistent results. We therefore studied the association between endothelial dysfunction and CVD under diverse circumstances. Methods and results Literature‐based meta‐analysis of prospective observational studies with ≥ 12 months of follow‐up published in Medline and having information on endothelial function and CVD outcomes. Tabular data on participant characteristics, endothelial function assessments and incident CVD outcomes were ed from individual studies. Random‐effects meta‐analysis was used to quantify pooled associations, and I2 statistic to evaluate between‐study heterogeneity. Potential sources of heterogeneity were explored by subgroup analyses and meta‐regression. Thirty five studies involving 17,206 participants met the inclusion criteria. During more than 80,000 person‐years of observation, up to 2755 CVD events were accrued, yielding a pooled relative risk (RR) of 1.25 (95% confidence interval 1.15–1.35) for CVD comparing top (i.e. more severe) vs. bottom (less severe) third of endothelial dysfunction. There was significant between‐study heterogeneity and evidence of publication bias. RRs varied importantly according to the method used to ascertain endothelial function, and were higher among older individuals and among participants with risk factors for CVD or established CVD at baseline. Conclusions Although endothelial dysfunction is an important determinant of cardiovascular outcomes in people with pre‐existing CVD, current evidence base does not support its use as a potentially useful measurement for risk stratification in people at lower risk of CVD. Linked Comment: Ferguson and Sattar. Int J Clin Pract 2015; 69: 624–5.