The present work investigates the relationship between semantic and prosodic (metric) processing in spoken language under 2 attentional conditions (semantic and metric tasks) by analyzing both behavioral and event-related potential (ERP) data. Participants listened to short sentences ending in semantically and/or metrically congruous or incongruous trisyllabic words. In the metric task, ERP data showed that metrically incongruous words elicited both larger early negative and late positive components than metrically congruous words, thereby demonstrating the online processing of the metric structure of words. Moreover, in the semantic task, metrically incongruous words also elicited an early negative component with similar latency and scalp distribution as the classical N400 component. This finding highlights the automaticity of metrical structure processing. Moreover, it demonstrates that violations of a word's metric structure may hinder lexical access and word comprehension. This interpretation is supported by the behavioral data showing that participants made more errors for semantically congruous but metrically incongruous words when they were attending to the semantic aspects of the sentence. Finally, the finding of larger N400 components to semantically incongruous than congruous words, in both the semantic and metric tasks, suggests that the N400 component reflects automatic aspects of semantic processing.