Prior studies have shown that many incidental features of survey questions create pragmatic effects-inferences about the meaning of the questions that lead to unwanted measurement error. As cooperative communicators who observe Grice's Cooperative Principle and its associated maxims, survey respondents sometimes read between lines and use every piece of information (including purely formal features like the numbers assigned to the response categories) to understand and answer survey questions. This study examines the impact of one formal feature of web survey questions-the physical arrangement of the questions on web pages-on survey respondents' inferences and responses. We presented a set of questions in a web survey either 1) one question per screen, 2) all on the same screen, or 3), in a grid on a single screen. I hypothesized that if respondents applied the maxim of relation during the survey response process, they would infer from the use of a grid that these seemingly irrelevant questions were related to each other. As a result of such a 'relatedness' inference, we would observe higher intercorrelations among the questions when they were in a grid. By contrast, the 'relatedness' inference would be weaker when question items were presented one question at a time on separate screens. To dampen and strengthen the possible effects of the physical layout of the questions, I varied the introduction to the items. One introduction indicated that the questions were taken from the same source (thus, encouraging respondents to apply the relation maxim); a second introduction indicated the items came from different sources so as to discourage the application of the relation maxim. Implications for using grids and instructions in designing web surveys will be discussed.