Thank you so much Francois!
I think I understand now this statistical significance in source space, at least at the basic level.
If I have two conditions, and if I have EPs, ERPs or frequency spectra related to both of these conditions, and if I want to see statistically significant difference between these two conditions, I should use statistical test.
On the other hand, if I have some ERPs of one condition, the normalization of sources with baseline gives me possibility to present only those ERP sources that are significantly different that the sources that explain the baseline activity.