Whether lexical selection is by competition is the subject of current debate in studies of monolingual language production. Here, I consider whether extant data from bilinguals can inform this debate. In bilinguals, theories that accept the notion of lexical selection by competition are divided between those positing competition among all lexical nodes vs. those that restrict competition to nodes in the target language only. An alternative view rejects selection by competition altogether, putting the locus of selection in a phonological output buffer, where some potential responses are easier to exclude than others. These theories make contrasting predictions about how quickly bilinguals should name pictures when non-target responses are activated. In Part 1, I establish the empirical facts for which any successful theory must account. In Part 2, I evaluate how well each theory accounts for the data. I argue that the data do not support theories that reject lexical selection by competition, and that although theories where competition for selection is restricted to the target language can be altered to fit the data, doing so would fundamentally undermine the distinctness of their position. Theories where selection is by competition throughout both target and non-target language lexicons must also be modified to account for the data, but these modifications are relatively peripheral to the theoretical impetus of the model. Throughout, I identify areas where our empirical facts are sparse, weak, or absent, and propose additional experiments that should help to further establish how lexical selection works, in both monolinguals and bilinguals.