As software ages, its quality degrades, leading to pathologies such as architectural decay, a problem that is hard to fight against. First, it is a slow drifting process that results from the natural evolution of software. Second, correcting it is a costly and dangerous operation that has the potential to affect the entire code base. A daring prospect on systems of multi-million lines of code. This spawned a lot of research, for example to measure the quality of a software archi-tectures. Yet a fundamental issue remains, we have very little means of measuring the quality of a given architecture. One of the reasons for this seems to be that building a formal experiment to test these issues is difficult due to the fuzzy nature of what is a good architecture , and to the need to use large and real systems. In this paper we setup an experimental design with well know systems such as Eclipse or JHotDraw that would allow to test the relevance of architectural design metrics. We illustrate it on some well known metrics.