One hundred forty-four scalp ictal EEGs from 54 patients were analyzed independently by three electroencephalographers for side and lobe of seizure onset. Observers did not know the patients' identities. Accuracy was determined by depth EEG. We found 58 to 60% agreement between observers for lobe, and 64 to 74% for side, of seizure onset; 21 to 38% agreement with depth EEG for lobe, and 46 to 49% agreement for side, of seizure onset; best accuracy for lateralization of seizure onset in temporal lobe seizures, but erroneous in 3 to 17%. More formal criteria are needed before scalp ictal records can be used reliably or accurately for localization.