Current views of lexical selection in language production differ in whether they assume lexical selection by competition or not. To account for recent data with the picture-word interference (PWI) task, both views need to be supplemented with assumptions about the control processes that block distractor naming. In this paper, we propose that such control is achieved by the verbal self-monitor. If monitoring is involved in the PWI task, performance in this task should be affected by variables that influence monitoring such as lexicality, lexicality of context, and time pressure. Indeed, pictures were named more quickly when the distractor was a pseudoword than a word (Experiment 1), which reversed in a context of pseudoword items (Experiment 3). Additionally, under time pressure, participants frequently named the distractor instead of the picture, suggesting that the monitor failed to exclude the distractor response. Such errors occurred more often with word than pseudoword distractors (Experiment 2); however, the effect flipped around in a pseudoword context (Experiment 4). Our findings argue for the role of the monitoring system in lexical selection. Implications for competitive and non-competitive models are discussed. (Contains 3 tables.)