BACKGROUND: Previous research suggests a link between antidepressant use and diabetes, but it is unclear whether the association is causal or attributable to detection/ascertainment bias. To examine this, we assessed the associations of antidepressant use with change in glucose levels and incidence of undiagnosed and diagnosed diabetes. METHODS: During an 18-year period, we monitored antidepressant use, glucose levels, and diabetes status in 5978 civil servants (70.9% male, age range 39-64 years) free of diabetes at baseline (the Whitehall II study). Use of medication and plasma glucose were assessed at four study screenings: 1991/1993, 1997/1999, 2003/2004, and 2008/2009. Incident diabetes cases were classified as either diagnosed (n = 294) if detected using self-report of physician diagnosis and/or the use of diabetes medication or undiagnosed (n = 346) if detected based on fasting and/or 2-hour postload glucose levels using an oral glucose tolerance test at the study screenings. RESULTS: Incidence of diagnosed diabetes was higher among antidepressant users than nonusers (odds ratio 3.10, 95% confidence interval: 1.66-5.78). However, antidepressant use was not associated with undiagnosed diabetes at any follow-up examination nor with higher fasting or 2-hour postload plasma glucose levels or increasing glucose levels over time. Odds ratio for undiagnosed diabetes for antidepressant users versus nonusers was .88 (95% confidence interval: .45-1.72, p = .70). The mean difference in glucose changes between participants reporting antidepressant use at three screenings compared with those not on antidepressant treatment was .0 mmol/L. CONCLUSIONS: The link between antidepressant use and diabetes risk may not be causal in nature.