The present study examines the decisions made by reasoners when they are asked to revise their beliefs in the face of new, counterfactual information. Participants indicated the Scope (the degree of set inclusion) of semantic generalizations about real categories in a Pretest. In subsequent experiments, these Scope values were used to predict the willingness of participants to retain statements in their existing knowledge sets. When those sets were logically compatible with a Modus Tollens (MT) structure, participants were more likely to retain the general statements, but not when the sets were logically compatible with a Modus Ponens (MP) structure. However, the MP retention rates increased when locatives were added to the generalizations. These findings are inconsistent with several prevailing proposals of belief revision but do support the concept of belief revision as following Possible Worlds logic.