What patterns of multimodal rhetoric characterizes the most successful texts meant for learning--inside or outside school? This article argues that the balance between cohesion and tension in the interplay between modes is an issue at the core of this challenging question. Texts meant for learning need both cohesion and tension in order to be both understandable and engaging. While cohesion has long been a central issue of interest among researchers of multimodal texts, the same can hardly be said about multimodal tension. The article offers a discussion of the concepts of multimodal cohesion and tension, relating them to three dimensions of user-text interaction: the dimensions of perception, interpretation and performativity. This elaboration concludes with the outline of a framework model for analytical purposes. The model is, thereafter, used in an illustrative analysis of an award-winning multimodal feature story collected from the online site of the "New York Times". (Contains 19 notes and 4 figures.)