Empirical software engineering focuses on the evaluation of software engineering technologies, such as processes and tools, by comparing related sets of data. It has contributed a valuable body of knowledge in several areas such as Software Economics and Software Quality, which in turn drove important advances in related tools and techniques. Unfortunately this is not (yet) the case for software architecture, where empirical studies are still few. Such a condition demands for further empirical research efforts on the topic of software architecture and suggests specific areas of improvement. In this paper we discuss several essential, innovative, and maybe provocative, questions such as: Why do we have so few applications of empirical software engineering on software architecture? Which are the main difficulties? What can we do?