We have produced a method to estimate ictal localized epileptic activity hidden among the background in scalp EEGs. When the visually completely different waveforms of the epileptic and background activities are nearly orthogonal, epileptic activity may be approximately extracted from the EEG data matrix by singular value decomposition with subsequent orthogonal rotation to match the distribution of one component with that of the epileptic source. A simulation study was carried out using a matrix mimicking the scalp EEG with an inconspicuous ictal epileptic activity from a dipole source. This hidden epileptic activity was approximately recovered by matching the dipole of interest with the epileptic dipole, even when the simulated waveforms of the epileptic and background activities were not exactly orthogonal. High linear correlation between these two types of waveforms hampered the recovery of the epileptic activity. In another simulation study employing two epileptic dipoles producing activities with the same waveform and a brief time lag, it was indicated that the temporal relationship between the epileptic activities could be also estimated using the cross-correlation function. In the preliminary clinical application of this method to the ictal EEGs of complex partial seizures, rhythmic activities with seemingly epileptic waveforms were estimated at the dipoles which were located in the vicinity of cortical lesions revealed by neuroimaging studies. These activities were indicated to appear before any change in the scalp EEG. We hope for the clinical application of this method for noninvasive estimation of inconspicuous ictal epileptic activity.