Generally, user modeling concerns a person interacting with a standing console. This scenario does not represent the +\SHU$XGLR system in use: a visitor freely moves in a museum, gathering information from an adaptive and portable electronic guide. To provide designers with presumptive user behavior, data about visitor profiles and visit styles were collected through a questionnaire. The study pointed out unpredicted situations (e.g., the importance of social context) and confirmed some working hypotheses (e.g., the relevance of visit span). This paper reports on this experience, describing how to go from designer questions to guidelines for user modeling, making the best use of empirical data.