Anecdotal and biographical reports have long suggested that bipolar disorder is more common in people with exceptional cognitive or creative ability. Epidemiological evidence for such a link is sparse. We investigated the relationship between intelligence and subsequent risk of hospitalisation for bipolar disorder in a prospective cohort study of 1,049,607 Swedish men. Intelligence was measured on conscription for military service at a mean age of 18.3 years and data on psychiatric hospital admissions over a mean follow-up period of 22.6 years was obtained from national records. Risk of hospitalisation with any form of bipolar disorder fell in a stepwise manner as intelligence increased (P for linear trend <0.0001). However, when we restricted analyses to men with no psychiatric comorbidity, there was a 'reversed-J' shaped association: men with the lowest intelligence had the greatest risk of being admitted with pure bipolar disorder, but risk was also elevated among men with the highest intelligence (P for quadratic trend=0.03), primarily in those with the highest verbal (P for quadratic trend=0.009) or technical ability (P for quadratic trend <0.0001). At least in men, high intelligence may indeed be a risk factor for bipolar disorder, but only in the minority of cases who have the disorder in a pure form with no psychiatric comorbidity.