Dynamic inconsistency reflects a prediction bias where decision makers fail to follow their plans simply because they experience outcomes on which their plan was based. Specifically, after experiencing an anticipated gain in one gamble, decision makers reject a second gamble they had planned to accept. The opposite pattern is found with losses. A common account of these findings is that prior outcomes are segregated during planned choices and are integrated only after being experienced. According to a derived “computational” hypothesis, integration of prior outcomes at the planning stage should reduce dynamic inconsistency while segregation should increase it. A “descriptive” meaning of segregation and integration offers the opposite hypothesis. An experiment that framed planned choices to encourage either integration or segregation of prior outcomes indicated that dynamic inconsistency persists in both framing conditions. We suggest alternative explanations for dynamic inconsistency, and discuss the difficulty of bridging between predicted and actual preferences. Copyright © 2005 John Wiley & Sons, Ltd.