When we assume that a survey reveals respondents’ true attitudes we tacitly assume that the subjects understood what we are asking them about and that they had no incentive to be untruthful. In typical studies none of the two assumptions holds. Subjects are asked questions that use undefined terms and they are asked about issues they have no incentive to answer truthfully. Here we argue that a way to solve the two problems lies in constructing a formal theory of an attitude in such a way that an attitude can be derived from the answers yet when answering the questions subjects cannot possibly know that their responses reveal anything about their attitude and, hence, they have no incentive to answer insincerely.We briefly discuss a study that has the desirable properties necessary for the proper design.