This study investigated the role of verbal working memory on bilingual lexical disambiguation. Spanish–English bilinguals read sentences that ended in either a cognate or noncognate homonym or a control word. Participants decided whether follow-up target words were related in meaning to the sentences. On critical trials, sentences biased the subordinate meaning of a homonym and were followed by targets related to the dominant meaning. Bilinguals with high span were faster at rejecting unrelated targets when the sentences ended in a homonym, whereas bilinguals with low span were slower. Furthermore, error rates for bilinguals with low span showed cognate inhibition, while bilinguals with high span showed no effects of cross-language activation. Results demonstrated that bilinguals with high span benefit from shared lexical codes whether these converge on to a single semantic representation (cognates) or not (homonyms). Conversely, bilinguals with low span showed inhibition from the competing lexical codes, even when they converge onto a single semantic representation.