INTRODUCTION: Heart rate variability (HRV) is a predictor of outcome in acute myocardial infarction and head trauma. Its efficacy in predicting outcome in stroke has not been well documented. MATERIALS AND METHODS: Twenty-five patients (mean age 39 years) with acute stroke treated in a stroke intensive care unit were studied. Continuous echocardiogram recorded for a 1-hour period was digitized and stored for off-line analysis. Time and frequency domain HRV measures were derived for the filtered and rectified ECG data for each patient. Clinical and HRV profiles were compared among patients who died or survived. RESULTS: At admission, 16 patients were comatose (Glasgow Coma score<9 at admission), 16 had focal weakness, and all had abnormal brain computed tomography. Of the 25 patients,11 died, 10 had a poor outcome, and 4 had good outcome. Two variables low-frequency (LF) spectral power and very low-frequency (VLF) spectral power correlated with mortality. After adjustment for mechanical ventilation and vasopressor administration, LF, VLF, and Triangular index of RR interval (TINN) correlated with mortality. On multiple regression analysis weighed for mechanical ventilation and vasopressor administration, the eye-opening score on Glasgow Coma Scale and LF spectral power were factors that were independently predictive of mortality. CONCLUSION: HRV measurements are independent predictors of outcome in acute severe stroke.