In this paper we contrast three implementations of Independent Component Analysis (ICA) as applied to epileptic scalp electroencephalographic (EEG) recordings, these are; Spatial (Ensemble) ICA, Temporal (single-channel) ICA and Spatio-Temporal ICA. These techniques are based on information derived from both multi-channel as well as single channel biomedical signal recordings. We assess the suitability of the three techniques in isolating and extracting out epileptic seizure sources. Although our results are preliminary in nature, we show that standard implementations of ICA (ensemble ICA) are lacking when attempting to extract complex underlying activity such as ictal activity in the EEG. Temporal ICA performs well in separating underlying sources, although it is clearly lacking in spatial information. Spatio-Temporal ICA has the advantage of using temporal information to inform the ICA process, aided by the spatial information inherent in multi-channel recordings. This work is being expanded for seizure onset analysis through scalp EEG.