The phonological loop model for retention of auditory verbal material in working memory, developed by Baddeley, assumes that irrelevant speech and phonological similarity influence only one and the same element of the system--that is, the phonological short-term store. We tested this idea by recording event-related potentials (ERPs) to auditorily presented letters that were phonologically similar or dissimilar and were to be memorized in the presence of more or less disturbing irrelevant speech. Irrelevant speech and phonological similarity caused ERP effects with clearly different scalp topographies, indicating that these factors influence different brain systems and hence probably different cognitive elements. Moreover, ERPs indicated that the phonological similarity effect might involve processes at the level of phonological analysis. Our data also support recent suggestions that the irrelevant speech effect is not based on the phonological similarity between relevant and irrelevant material, but on the phonological variability within the irrelevant stream.