High-frequency words are often assumed to be the most useful words for communication, as they provide the greatest coverage of texts. However, the relationship between coverage and comprehension may not be straightforward; how words relate semantically to people’s mental representations is also important. In this study, we evaluate how useful different sets of ‘core vocabularies’ are in text comprehension. The core vocabularies, which reflect different aspects of distributional and semantic information, provide different amounts of information for different vocabulary size and amount of text coverage. In our experiment, we showed people narrative texts with all but the core words removed, and measured comprehension in a variety of ways. Our results show that both distributional (e.g., frequency-based) and semantic (e.g., word association-based) core vocabularies are communicatively useful, but that the semantically-based core vocabularies provide more information when textual coverage is held constant.