The effects of working memory (WM) on the semantic N400 response were studied using high-resolution event-related brain potentials (ERPs). Participants were presented with semantically related sentence pairs and the terminal word congruence was varied in the second sentence. WM load was varied for the sentence pairs using a modified fan procedure [J.R. Anderson, Retrieval of propositional information from long-term memory. Cogn. Psychol., 6 (1974) 451–474; J.R. Anderson, A spreading activation theory of memory. J. Verbal Learn. Verbal Behav., 22 (1983) 261–295]. ERPs were recorded to the onset of the terminal word (Congruent and Incongruent) for sentence pairs at two levels of WM load (WML1 and WML2). Behavioral data analysis revealed that reaction times (RTs) increased as a function of WM load (i.e., the fan effect). Semantically incongruent words elicited an N400 response. Increased WM load reduced the congruency effect and, importantly, resulted in a significant delay in N400 peak latency (∼50 ms). Moreover, the latency delay was correlated with a neuropsychological measure of individual WM capacity. WM load effects on the N400 were interpreted as a result of competing activation in WM, both modulating semantic expectancies and delaying semantic integration. Exploratory source analysis revealed activation in occipital, temporal, and parietal regions. Consistent with prior reports on the N400 and semantic processing, equivalent current dipoles were modeled in the left perisylvian region. Increased WM load led to novel source activation in the left inferior parietal region as well as increased activation levels in anterior temporal sources. The source modeling results were in agreement with the functional roles typically ascribed to these areas and confirmed that the scalp-recorded WM load effects on the N400 were present within the intracranial generators.