Answers to attitude questions can vary sharply as a function of earlier items in the questionnaire. Some of these context effects involve the retrieval of beliefs relevant to the question. These beliefs are often numerous and complex, but the retrieval process is not usually very thorough or reliable. Many respondents retrieve only a sample of their beliefs about an issue, a sample that overrepresents whatever beliefs happen to be most accessible at the time the question is asked. Which beliefs are most accessible is affected by prior questions on related topics. We explored the effects of item context in an experiment. Respondents in a telephone survey were interviewed twice about legalized abortion, welfare spending, aid to the Contras, and the proposed “Star Wars” defense system; in the second interview, the items that preceded the target item on each issue were varied systematically. For example, some respondents were asked several questions about the U.S. role in the Vietnam War before they were asked about policies toward Nicaragua; others were questioned about Cuba before being asked about Nicaragua. The preceding items affected responses to the target items on three of the four issues. Two other findings support an explanation of these context effects in terms of belief accessibility: The effects of context were especially marked when the context items were presented in a block immediately prior to the target item and when the context and target items were closely related.