Word learning is guided by the statistical co-occurrence between spoken words and potential referents, through which learners gradually map labels to objects across situations. Given that word learning does not occur in a vacuum, rather in a communicative context, it is relevant to evaluate the role that speakers play. Because we do not evaluate the information provided by every person equally, it is reasonable to think that someone who makes lexical errors is not a reliable speaker from whom to learn new words. The current study focuses on speaker reliability in adult cross-situational word learning (CSWL). In two experiments we investigated the extent to which adults attend to the reliability of the speaker and how this affects word learning in a CSWL task. We varied the consistency with which a speaker mapped novel words to familiar objects. We hypothesized (1) that the speakers’ reliability would be judged differently depending on their past object-labeling accuracy, and (2) that new words would be more difficult to learn when presented by an unreliable speaker. Experiment 1 shows that the unreliable speaker was assessed as less reliable, compared to the reliable speaker, but this effect disappeared in Experiment 2, when participants were taught new words by two speakers, a reliable and an unreliable one. Furthermore, we found no evidence to support the hypothesis that being exposed to an unreliable speaker impairs CSWL in adults. We discuss the relevance of these findings and the importance of further research on the role of speaker reliability in CSWL.