Gaining an understanding of software systems is an important discipline in many software engineering contexts. It is essential that software engineers are assisted as much as possible during this task, e.g., by using tools and techniques that provide architectural views on the software at hand. This Ph.D. research addresses this issue by employing dynamic analysis for the reconstruction of such views from running systems. The aim is to devise new abstraction techniques and novel visualizations, to combine them, and to assess the benefits through substantial case studies and controlled experiments. This paper describes our approach, reports on the results thus far, and outlines our future steps.