Responses to survey questions are assumed to be based on what respondents can actually draw from memory. For example, if respondents report engaging in some behavior a certain number of times, we assume that a substantive answer reflects what they really know. Conversely, when respondents say that they "don't know" (DK) the answer, it is believed that their response reflects a lack of memory relevant to the question. However, survey responding is now understood to be a highly complex process. In this article, we propose a framework to account for the kind of reports made by respondents to a survey. The framework holds that respondent reports are based on three factors: what the respondent actually knows (the person's "cognitive state"); whether the respondent believes that the potential answer meets the requirements of the question (an "adequacy judgement"); and whether the respondent chooses to provide this answer (the person's "communicative intent"). Subsequently, the article examines how relevant findings in the literature support the framework, and then evaluates the assumptions made about cognitive states in light of the results of two experiments. The literature and research presented demonstrate that survey responding can be accounted for by a framework that assumes responding is based on certain key cognitive processes.