How does human language comprehension achieve its hallmark ability to construct an unbounded range of compositional meanings from a finite (though large) set of individual words? My lab uses event-related brain potentials (ERPs) to investigate the neuro-cognitive processes serving on-line sentence comprehension. ERPs provide sensitivity to distinct neural processes with millisecond-level temporal resolution and a limited ability to resolve neuroanatomical sources. Recent findings in our lab and others’ indicate that semantic knowledge can drive combinatory interpretations somewhat independently of syntactic analysis, sometimes overwhelming opposing, unambiguous syntactic cues. Our findings also indicate that semantic knowledge operates in a predictive manner, allowing semantic influences on the earliest aspects of the brain’s response to an incoming word, well within the initial ~200 msec of the onset of the visual stimulus during reading. I will discuss the implications of these findings for psycholinguistic models, including classical syntax-first proposals and recent proposals that emphasize the contributions of semantic knowledge systems to the computation of combinatory language interpretation.