In a language processing experiment, two different types of argument structure violations in German were tested (Frisch, 2000, Experiment 4). As expected, both violations elicited a biphasic N400-P600 response in event-related brain potentials (ERPs) compared to correct sentences (see Friederici & Frisch, 2000). However, the customary voltage averaging technique was unable to reveal a theoretically predicted difference in the N400 component between the two violation conditions. Our analysis rests on a coarse-graining of the ERP epochs by assigning different symbols to samples depending on whether the signal is above or below some threshold. Thus, only polarity and latency information are considered (beim Graben et al., 2000). Here, we employed a three-symbol static encoding with varying thresholds, followed by a spin-flip transformation as a nonlinear filter. Finally we computed an estimator of the signal-to-noise ratio (SNR) for the symbolic dynamics (beim Graben, 2001) measuring the coherence of threshold-crossing events. Hence, we utilized the inherent noise of the EEG for sweeping the underlying ERP components beyond the threshold, i.e. stochastic resonance (beim Graben & Kurths, 2003). Drawing the SNR computed within the time windows of the ERP components against the encoding thresholds yields characteristic resonance curves for each component. With respect to the present experiment, we found differences in the resonance curves for the two N400 components, thus indicating amplitude differences between the two N400s across single trials. In sum, we have shown that SRA is a promising method complementary to customary averaging analysis since it is more sensitive with respect to experimentally interesting differences which traditional techniques are unable to uncover.