This article addresses the questions of how and when lexical information influences phoneme identification in a series of phoneme-monitoring experiments in which conflicting predictions of autonomous and interactive models were evaluated. Strong facilitatory lexical effects (reflected by large differences in detection latencies to targets in words and matched nonwords) were found only when targets came after the uniqueness point of the target-bearing word. Furthermore, no evidence was obtained for lexically mediated inhibition on phoneme identification as predicted by the interactive activation model TRACE. These results taken together point to strong limitations in the way in which lexical information can affect the perception of unambiguous speech.