Intelligent Tutoring System are state of the art in eLearning since the late 1980s. The earliest system have been developed in teams of psychologists and computer scientists, with the goal to investigate learning processes and, later on with the goal to intelligently support teaching and training with computers. Over the years, the eLearning hype appeared and vanished, and returned in a slightly different shape. However, Intelligent Tutoring Systems, as one of the most adaptive system types, remained in the focus of interest. Due to this, it is surprising that there exists no ITS architecture, no guideline how to implement these systems, and no structured, software engineered approach, which helps to develop comparable systems. In this paper, we will show the first steps toward such a system: a broad analysis of ITS over several years of research, with the resulting structured architecture. [For full proceedings, see ED571332.]