Two experiments are reported that test the hypothesis that a reader can make use of the size of the semantic domain activated by a sentence context when inferring the meaning of a partially known word. We investigated words at three levels of knowledge: known, frontier, and unknown (e.g., Durso & Shore, 1991). Experiment I demonstrated that participants have knowledge about the meanings of words that they deny are part of the language (the unknown level), and that they make use of relative differences in the size of the semantic domains tapped by two sentences when asked to decide on correct usage of these unknown words. Experiment 2 demonstrated that participants have knowledge about the general semantic constraints operating on their unknown words, even when relative differences in size of semantic domains are controlled. Implications for the role of contextual constraints in vocabulary acquisition are discussed.