In the recent years, Software Engineering research has shown the rise of interest in the empirical studies. Such studies are often based on empirical evidence derived from corpora - collections of software artifacts. While there are established forms of carrying out empirical research (experiments, case studies, surveys, etc.), the common task of preparing the underlying collection of software artifacts is typically addressed in ad hoc manner. In this thesis, by means of a literature survey we show how frequently software engineering research employs software corpora and using a developed classification scheme we discuss their characteristics. Addressing the lack of methodology, we suggest a method of corpus (re-)engineering and apply it to an existing collection of Java projects. We report two extensive empirical studies, where we perform a broad and diverse range of analyses on the language for privacy preferences (P3P) and on object-oriented application programming interfaces (APIs). In both cases, we are driven by the data at hand, by the corpus itself, discovering the actual usage of the languages.