This study examines, within the domain of science, the characteristics of words that predict word knowledge and word learning. The authors identified a set of word characteristics--length, part of speech, polysemy, frequency, morphological frequency, domain specificity, and concreteness--that, based on earlier research, were prime candidates to explain variation in word knowledge and word learning. The outcome measures were the pretest (evidence of word knowledge) and posttest (evidence of word learning) vocabulary scores of second- through fourth-grade students who participated in one of several studies designed to evaluate the efficacy of science units that were part of a multiyear research and development program for an integrated science and literacy curriculum. The authors first examined individual predictors and then built multivariate models from the individually significant predictors of pretest score. Three characteristics were predictive of word knowledge (pretest score) at two or more grade levels; frequency, polysemy, and length predicted word difficulty independent of instruction. These three characteristics accounted for 39% of the variance in third graders' pretest scores. Polysemy and frequency alone accounted for 34% of the variance at second grade and 23% at fourth grade. In addition, frequency and polysemy explained students' vocabulary growth scores (posttest controlling for pretest) over the course of instruction at two of three grade levels. Understanding which characteristics of words are related to difficulty within the domain of science may provide a principled basis both for the selection of words for instruction and for differentiating instruction across categories of science words.