Comprehenders can rapidly use both their linguistic knowledge and different kinds of information in visual context during language comprehension. Little is known, however, about the relative time courses and mechanisms by which different kinds of visual information influence language comprehension. We recorded event-related brain potentials (ERPs) as participants read a subject-verb-object sentence and verified whether or not it matched different (verb-action versus role relations) aspects of a recently viewed picture. When the verb- action did not match the depicted action, we replicated larger N400s (300-500ms) over centro-parietal scalp to the verb (300- 500 ms) relative to the responses for matches. In contrast, ERP effects to role-relation mismatches (a person depicted as undergoing an action but described as performing it) qualitatively differed from and occurred prior to the verb-action congruence N400. Our findings implicate at least two temporally distinct mechanisms governing picture-sentence verification processes.