Blind source separation (BSS) is an emerging statistical and data processing technique which aims to recover unobservable source signals from the observed mixtures. Second-order blind identification (SOBI) is one BSS algorithm that relies on stationary second-order statistics based on joint diagonalization of a set of covariance matrices. In simulations, the use of multiple covariance matrices computed with different time delays, τs, was beneficial for source separation, particularly when the underlying sources had highly overlapping spectra. Given the spectral overlap between actual brain sources, we experimented with different sets of temporal delays to empirically determine their effects on the isolation of electrical signals arising from a temporally and spatially well characterized brain location, the primary somatosensory cortex (SI). Using EEG data collected during median nerve stimulation, we found that the successful isolation of left and right SI activity required the use of a range of time delays and that the best separation was observed when the largest range of τs from 1 up to 300 ms was used.