BACKGROUND: Although decreased heart rate variability (HRV) is an independent predictor of death in various populations, its prognostic value in patients with end-stage renal disease on chronic haemodialysis is unknown. METHODS: We prospectively studied 120 chronic haemodialysis patients (age 61+/-11 years; males 51%; diabetics 38%; duration of haemodialysis therapy 50+/-114 months) who underwent 24 h electrocardiography at baseline for analysis of time- and frequency-domain HRV. RESULTS: All HRV measures in the patients were significantly reduced compared with those obtained from 62 age-matched healthy subjects. During a follow-up period of 26+/-10 months, 21 patients died (17.5%); 10 from cardiac causes and 11 from non-cardiac causes (seven fatal strokes and four other causes). A Cox proportional hazards model revealed that, of the HRV measures, decreases in the triangular index (TI), very-low-frequency (0.0033-0.04 Hz) power, ultra-low-frequency (<0.0033 Hz) power (ULF) and the ratio of low-frequency (0.04-0.15 Hz) power to high-frequency (0.15-0.4 Hz) power had significant predictive value for cardiac death. None of the HRV measures, however, had predictive value for non-cardiac death, including stroke death. Even after adjustment for other univariate predictors including age, diabetes, serum albumin and coronary artery disease, the predictive value of decreased TI and ULF remained significant-adjusted relative risk (95% confidence interval) per 1 SD decrement of TI and ULF, 3.28 (1.08-9.95) and 1.92 (1.01-3.67), respectively. CONCLUSIONS: Decreases in some HRV measures, particularly those reflecting long-term variability, are independent predictors of cardiac death in chronic haemodialysis patients.