Background: The predictive value of heart rate variability (HRV) in peritoneal dialysis (PD) has never been tested. Methods: In this study, the associations between HRV measures and the mortality in 81 PD patients were analyzed. HRV was measured by using 5-min recordings of a stationary system by a standardized method. Both time domain and frequency domain parameters were analyzed. Results: During a follow-up period of 43.78 +/- 14.77 months, 25 patients died, four patients were transferred to hemodialysis. Of the 81 patients, the time domain parameters, such as the standard deviation of differences between adjacent normal sinus to normal sinus (NN) intervals (SDSD) and the square root of the mean of the squared differences between adjacent normal NN intervals (RMSSD), were higher; the frequency domain parameters, such as the ratio of low-frequency power to high-frequency power (LF/HF) and the normalized LF, were lower, and the normalized HF was higher in the non-survived group as compared with the survived group. A Cox proportional hazards model analysis revealed that, of the HRV measures, decrease of the normalized LF, LF/HF and increase of rMSSD, SDSD, normalized HF had significant predictive value for mortality. After adjustment for other univariate predictors including age, urine volume, renal Kt/V, high-sensitivity C-reactive protein (hs-CRP), the predictive value of decreased LF/HF remained significant. Kaplan-Meier survival analysis showed mortality rate was much higher in patients with a low LF/HF (median value of 1.56). Conclusion: The decreases of LF/HF which reflects impaired sympathetic nerve regulation is an independent predictor of mortality in PD patients.