End users can ask themselves about the User Interface (UI). Questions arise because users are not designers so both designers and users, have different perceptions of the same UI. Help Systems have naturally emerged to tackle this problem. Most of these Help Systems are predefined, so at design time designers need to anticipate the problems users may find at runtime, which limits the scope of the support. This paper explores Model-Driven Engineering to overcome this limitation: models created at design time are exploited at runtime for providing end users with explanations. Based on Norman’s Theory of Action this paper introduces the Gulf of Quality as the distance between the models the designer creates at design time and the mental models the end user elaborates. This concept sets the basis of a Model-Driven method and a supporting architecture for computing explanations for the end user. The method deals uniformly with the UI of the help system and the UI of the application. They can be weaved or not, depending on the model transformations the designer selects. A software architecture is devised and implemented in a running IDE. The feasibility of the approach is shown through two use cases.