INTRODUCTION: Short-term heart rate variability (HRV) may change immediately before onset of a ventricular arrhythmia (ONSET). METHODS AND RESULTS: Power spectrum analysis was performed on instantaneous heart rate (IHR; including all beats) and normal heart rate (NHR; excluding ectopics) curves obtained at equally spaced 0.5-second intervals using a cubic spline. The database consisted of 135 sets of 1,024 RR intervals leading to ventricular arrhythmia (VA) and controls from 78 patients. Total periodogram and time course of spectral power were obtained. Ten spectral bands of 0.1-Hz bandwidth (0 to 1 Hz) were analyzed. A simple threshold technique was retrospectively used to predict the onset of a VA. RR intervals that led to VA ONSET had significantly higher total spectral power than controls (P < 0.001 for both NHR and IHR for every band). Spectral power remained constant until 100 seconds before ONSET and then increased significantly in the time window immediately preceding ONSET (P < 0.02 compared with others). Using a simple threshold method, a predictive accuracy of 68%+/-1.4% was obtained with different window sizes. Using specific spectral bands, the predictive accuracy of VA ONSET could be improved to 76% for IHR and 71% for NHR (0.8- to 0.9-Hz band). CONCLUSION: Our results suggest that a sustained higher power increase in NHR and IHR occurs during the course of 12.11+/-.57 minutes, followed by a sudden elevation in spectral power within 100 seconds of ONSET, and may be a precursor to ventricular tachycardia/ventricular fibrillation episodes.