Aim Previous research has indicated an association between diabetes and anxiety. However, no synthesis has determined the direction of this association. The aim of this study was to determine the longitudinal relationship between anxiety and diabetes. Methods We searched seven databases for studies examining the longitudinal relationship between anxiety and diabetes. Two independent reviewers screened studies from a population aged 16 or older that examined either anxiety as a risk factor for incident diabetes or diabetes as a risk factor for incident anxiety. Studies that met eligibility criteria were put forward for data extraction and meta‐analysis. Results In total 14 studies (n = 1 760 800) that examined anxiety as a risk factor for incident diabetes and two (n = 88 109) that examined diabetes as a risk factor for incident anxiety were eligible for inclusion in the review. Only studies examining anxiety as a risk factor for incident diabetes were put forward for the meta‐analysis. The least adjusted (unadjusted or adjusted for age only) estimate indicated a significant association between baseline anxiety with incident diabetes (odds ratio 1.47, 1.23–1.75). Furthermore, most‐adjusted analyses indicated a significant association between baseline anxiety and incident diabetes. Included studies that examined diabetes to incident anxiety found no association. Conclusions There was an association between baseline anxiety and incident diabetes. The results also indicate the need for more research to examine the direction of association from diabetes to incident anxiety. This work adds to the growing body of evidence that poor mental health increases the risk of developing diabetes. What's new? This is the first synthesis to determine the direction of association between anxiety and diabetes. The results indicate that anxiety is a risk factor for incident diabetes. The results also indicate that there is a need for further research into diabetes as a risk factor for incident anxiety.