Epilepsy affects over 50 million people worldwide, and it is characterized by seizures that, when uncontrolled, may cause serious injuries to patients and people around them. An important proportion of epileptic patients responds poorly to medication or does not receive adequate treatment, so a seizure prediction system could contribute to improving their quality of life. Current seizure prediction approaches are mainly based on the use of video-electroencephalography (EEG), which may be unrealistic for home and ambulatory monitoring. Since abnormal epileptic activity in the brain also can affect brain regions controlling autonomic activity, fluctuations in cardiac activity may reflect such anomalies. Previous studies report significant differences between time- and frequency-domain heart rate variability (HRV) measures calculated near and far from seizure onsets, thereby suggesting that HRV could be useful for predicting seizures. However, those differences were observed around 3-30 min before seizure onsets, thereby suggesting that patients have to wait for the same time before determining whether a seizure will occur or not. As an alternative approach, this work aims to examine the feasibility of using ultra-short-term HRV (i.e., 1-min ECG recordings) for seizure prediction. ECG recordings from seven epileptic patients were used, and time- and frequency-domain HRV measures were extracted from twenty 1-min length segments previous to seizures onset. RMSSD, SDNN/RMSSD, and LF mean values showed significant differences (p < 0.05) 1 minute before the seizure, which suggests that patients may have little time to ensure their safety. Future research endeavors may include exploring how suitable are non-linear HRV indexes as epileptic seizures predictors.