OBJECTIVES: Characterization of the EEG pattern during the early phase of a seizure is crucial for identifying the epileptic focus. The purpose of the present investigation was to evaluate a method that divides ictal EEG activity into segments of relatively constant surface voltage distribution, and to provide a 3-dimensional localization of the activity during the different segments. METHODS: For each timepoint the electrical voltage distribution on the scalp (the voltage map) was determined from the digitized EEG recording. Through a spatial cluster analysis time sequences where the maps did not change much (segments) were identified, and a 3-dimensional source reconstruction of the activity corresponding to the different mean maps was performed using a distributed linear inverse solution algorithm. RESULTS: Segments dominating early in seizure development were identified, and source reconstruction of the EEG activity corresponding to the maps of these segments yielded results which were consistent with the results from invasive recordings. In some cases a sequence of consecutive segments was obtained, which might reflect ictal propagation. CONCLUSIONS: Segmentation of ictal EEG with subsequent 3-dimensional source reconstruction is a useful method to non-invasively determine the initiation and perhaps also the spread of epileptiform activity in patients with epileptic seizures.