In this paper the trial-to-trial variability of brain responses (as opposed to the non-event-related background variability) in flash-evoked potential experiments is investigated by means of two new statistical tests, proposed by M ocks et al. (1984). In two groups of normal children, a considerable number of inhomogeneous responders could be detected, and evidence was found that two different modes of response variability are independently present in the data. The paper also gives a precise description of the computing algorithm and an empirical validation of the applicability to EP data of the two tests.