Counterfactual reasoning occurs when people are asked to assume for the sake of argument that a fact they previously thought was true is now false and to draw a conclusion on that basis. To accomplish this sort of reasoning requires a revising of one's beliefs, which was simulated in the present study. Students were shown a set of statements that they were to assure themselves was consistent. They were then asked to accept a counterfactual assumption as true and reconcile resulting inconsistencies among the set of statements. In these problems, one statement is a generality (e.g., All trees on the plaza are elms), another is a particular (e.g., This tree is a pine), and one is a counterfactual (e.g., Assume this tree is on the plaza). Students preferred to reconcile the inconsistency by identifying the generality as "true" and the particular as "false." They did this more often when the assumption combined categories than when it dislodged categories and when real beliefs were at stake rather than arbitrary generalities. This study tested current models of inference for their ability to account for counterfactual reasoning and found the results to be consistent with natural deduction system, mental models, and conceptual-integration network approaches to everyday reasoning.