An unsupervised correlation-based clustering method was developed to assess the trial-to-trial variability of auditory evoked potentials (AEPs). The method first decomposes single trials into three frequency bands, each containing activity primarily associated with one of the three major AEP components, i.e., P50, N100 and P200. Next, single-trial evoked potentials with similar post-stimulus characteristics are clustered and selectively averaged to determine the presence or absence of an AEP component. The method was evaluated on actual AEP and spontaneous EEG data collected from 25 healthy participants using a paradigm in which pairs of identical tones were presented, with the first stimulus (S1) presented 0.5s before the second stimulus (S2). Homogeneous, well-separated clusters were obtained and substantial AEP variability was found. Also, there was a trend for S2 to produce fewer 'complete' (and significantly smaller) responses than S1. Tests conducted on spontaneous EEG produced similar clusters as obtained from EP data, but significantly fewer stimuli produced responses containing all three EP components than seen in AEP data. These findings suggest that the clustering method presented here performs adequately to assess trial-to-trial EP variability. Also, the results suggest that the sensory gating observed in normal controls may be caused by the fact that the second stimulus generates fewer 'responsive' trials than the first stimulus, thus resulting in smaller ensemble averages.