As multimedia technologies largely increase communication effectiveness between humans and computers, the importance of efficient multimedia authoring tools commands the attention of both researchers and software venders. Many presentation or authoring tools were developed for presenters or artists in various fields. However, presentations created by these tools were either communicating with its addresses in a single direction, or providing limited navigation controls for the audiences via push buttons or menus. These presentations cannot incorporate addressees' responses. As a result, an audience watches the same demonstration over and over again even if he/she has told the computer he/she understands the topic. In this paper, we introduce a multimedia presentation design system that allows a presenter to plan the audience's reaction in advance. While the audience is watching a presentation, the underlying inference system is learning from his/her response. This mechanism makes a presentation to be proceeded again and acted according to the audience's background and knowledge. Thus, the resulting presentation is more diversified. In this paper, a model of presentation is defined. Based on this model, we design a graphical user interface and a language to facilitate the presentation design.