An experiment compared the impact of more and less semantically connected sentence contexts on vocabulary learning. Third graders (N = 40) were taught the definitions and meanings of six unfamiliar verbs: "anticipate," "attain," "devise," "restrain," "wield," and "persist." The verbs were embedded in four sets of sentences written to fill syntactic and semantic slots that fully activated the verbs' meanings. Students were randomly assigned to one of two conditions. In the cohesive condition, the verbs were embedded in sentences that contributed events to a common scenario such as a birthday party, hence creating semantic connections among the verbs and other content words in the sentences. The meanings of the verbs bore no inherent relationship to the scenarios, so connections arose from the sentence contexts. In the unconnected condition, the verbs were embedded in sentences that were equally meaningful but described independent, unrelated events and hence minimized semantic connections among the verbs and sentences. An experimenter read definitions and sentences orally while students followed along in print. The hypothesis tested was that students would learn verb meanings better in cohesive than unconnected contexts. This was confirmed on posttests assessing memory for definitions, sentence generation, and correct verb usage in sentences. Explanations of the effect of context cohesion on vocabulary learning are drawn from latent semantic analysis and schema theory. Word-reading skill was a stronger predictor of students' vocabulary learning than their general vocabulary knowledge despite no need to decode words, suggesting the contribution of orthographic mapping to vocabulary learning.