Objective: We implemented and tested an existing seizure detection algorithm for scalp EEG (sEEG) with the purpose of improving it to intracranial EEG (iEEG) recordings. Method: iEEG was obtained from 16 patients with focal epilepsy undergoing work up for resective epilepsy surgery. Each patient had 4 or 5 recorded seizures and 24 hours of non-ictal data were used for evaluation. Data from three electrodes placed at the ictal focus were used for the analysis. A wavelet based feature extraction algorithm delivered input to a support vector machine (SVM) classifier for distinction between ictal and non-ictal iEEG. We compare our results to a method published by Shoeb in 2004. While the original method on sEEG was optimal with the use of only four subbands in the wavelet analysis, we found that better seizure detection could be made if all subbands were used for iEEG. Results: When using the original implementation a sensitivity of 92.8% and a false positive ratio (FPR) of 0.93/h were obtained. Our extension of the algorithm rendered a 95.9% sensitivity and only 0.65 false detections per hour. Conclusion: Better seizure detection can be performed when the higher frequencies in the iEEG were included in the feature extraction. Our future work will concentrate on development of a method for identification of the most prominent nodes in the wavelet packets analysis for optimization of an automatic seizure detection algorithm.