Abstract Since the early days of generative grammar, the study of "unbounded dependencies" such as wh-questions and relative clauses has occupied a central place in both syntactic theory and language processing research. The problem that such constructions pose is as follows. In a normal wh-question, a wh-phrase is typically displaced to the left periphery of a clause (What did you say - to John?); this displaced constituent is often referred to as a "filler." The vacant position (indicated in the previous example by a blank line) where it would ordinarily occur in an "echo" question (You said what to John?) is correspondingly referred to as a "gap." Filler and gap are mutually dependent on each other since they share syntactic and semantic information essential for successful sentence interpretation. However, since sentence processing is a sequential operation, a filler cannot be assigned to its gap until some time after it has occurred. In other words, the filler must be held in working memory until such time as filler-gap assignment can take place. The intent of the research reported here was to examine the processing of unbounded dependencies in English as revealed in event-related brain potentials (ERPs). To this end, subjects were shown both grammatical and ungrammatical yes/no-questions (Did you say something to John?) and wh-questions. A number of comparisons made at various points in these questions showed that both the storage of a filler in working memory and its subsequent retrieval for filler-gap assignment were associated with an enhanced negativity between 300 and 500 msec poststimulus over left anterior sites. This effect of left anterior negativity (LAN) was independent of and orthogonal to the grammaticality of the eliciting condition. We show how this interpretation coincides with recent studies that demonstrate a correlation between left anterior negativity, working memory capacity, and successful language processing.