1 Purpose : This study proposes the use of a new classification algorithm, Reservoir Computing, to develop a real-time and accurate epileptic seizure detection system. 2 Methods: Reservoir Computing (RC) is a training method for recurrent neural networks where only a simple linear readout function is trained and where the neural network, the reservoir, is randomly created. As input for this reservoir we use a selection of different EEG features currently existing in seizure detection literature. This selection was made during training using a basic feature selection method. The output of the reservoir was trained using a ridge regression algorithm. 3 Results : In this study intracranial rat data from two different types of generalized epilepsy are detected: absence and tonic-clonic epilepsy. For both seizure types our approach resulted in an area under the Receiver Operating Characteristics curve (AUC) of 0.99 on the test data. For absences an average detection delay of 0.3s was noted, for tonic-clonic seizures this was 1.5s. The SWD detection method was tested on 15 hours of EEG-data coming from 13 GAERS rats, from which 10% was used for training. Our method outperformed the other implemented methods from which the best method was developed by Fanselow et al. in 2000 and resulted in an AUC of 0.96 and an average detection delay of more than 3 seconds. To evaluate the tonic-clonic seizure detection method 4 hours and 23 minutes of data of 4 rats was used. 20% of the total dataset was used for training, the rest was used for testing. Again our method outperformed other methods where the best method by White et al. in 2006 which resulted in a AUC of 0.82. 4 Conclusion : This study shows that it is possible to perform seizure detection using the described Reservoir Computing method and that it outperforms existing methods.