OBJECTIVES--To determine the incidence of a long QT interval as a marker for sudden death in patients with anorexia nervosa and to assess the effect of refeeding. To define a long QT interval by linear regression analysis and estimation of the upper limit of the confidence interval (95% CI) and to compare this with the commonly used Bazett rate correction formula. DESIGN--Prospective case control study. SETTING--Tertiary referral unit for eating disorders. SUBJECTS--41 consecutive patients with anorexia nervosa admitted over an 18 month period. 28 age and sex matched normal controls. MAIN OUTCOME MEASURES--maximum QT interval measured on 12 lead electrocardiograms. RESULTS--43.6% of the variability in the QT interval was explained by heart rate alone (p < 0.00001) and group analysis contributed a further 5.9% (p = 0.004). In 6 (15%) patients the QT interval was above the upper limit of the 95% CI for the prediction based on the control equation (NS). Two patients died suddenly; both had a QT interval at or above the upper limit of the 95% CI. In patients who reached their target weights the QT interval was significantly shorter (median 9.8 ms; p = 0.04) relative to the upper limit of the 60% CI of the control regression line, which best discriminated between patients and controls. The median Bazett rate corrected QT interval (QTc) in patients and controls was 435 v 405 ms.s-1/2 (p = 0.0004), and before and after refeeding it was 435 v 432 ms.s1/2 (NS). In 14(34%) patients and three (11%) controls the QTc was > 440 ms.s-1/2 (p = 0.053). CONCLUSIONS--The QT interval was longer in patients with anorexia nervosa than in age and sex matched controls, and there was a significant tendency to reversion to normal after refeeding. The Bazett rate correction formula overestimated the number of patients with QT prolongation and also did not show an improvement with refeeding.