Epilepsy is the second most common neurologic disorder behind stroke, affecting 1% of the world's population, and is particularly debilitating because of the unpredictability of epileptic seizures. A system to automatically detect or predict the onset of seizures would allow for a seizure warning system or for application of novel therapies to control or eliminate seizures. Since the 1970s, a variety of methods have been devised to detect and predict seizures with limited efficacy due to the variability in the electrographic nature of seizures both within and across subjects. In contrast to standard classification methods for seizure detection and prediction, in this project, we are investigating the use of a signal-detection algorithm called SIGFRIED (Signal modeling For Real-time Identification and Event Detection). Within this framework, we seek to model a patient's baseline electrographic condition using several hours of baseline activity and, through the SIGFRIED algorithm, determine the difference from this baseline condition. Initial work has focused primarily on frequency band-specific power changes as input features for SIGFRIED modeling. Currently, 590 hours of clinical data from 6 invasively monitored patients has been used to produce sensitivity of seizure detections greater than 90% with less than 3 false positives per hour. Continuing work on this project involves seeking improvement of seizure detection and prediction performance through fine tuning the seizure prediction algorithm, as well as through the use of additional input functions as features for determining the SIGFRIED model and novelty score.