In this study, we report our development of a patient-specific rule-based seizure prediction system. Five univariate and one bivariate nonlinear measures were extracted from non-overlapping 10-second segments of intracranial EEG (iEEG) data recorded using both depth electrodes in the brain and subdural electrodes over the cortical surface. Nonlinear features representing the specific characteristic properties of EEG signal were then integrated spatio-temporally in a way to predict to predict seizure with high sensitivity. The present system was tested on 58 hours of iEEG data containing ten seizures recorded in two patients with medically intractable focal epilepsy. Within a prediction horizon of 30 and 60 minutes, our method showed an average sensitivity of 90% and 96.5% with an average false prediction rate of 0.06/h and 0.055/h, respectively. The present results suggest that such a rule-based system can become potentially a useful approach for predicting seizures prior to onset.