In this thesis, I present a class abstraction technique (CAT) that supports the testing process by capturing aspects of software complexity based on the combination of class characteristics present in Java applications. I describe TaxTOOLJ, which is the tool that was developed to catalog Java classes based on this CAT, and detail the experiments that were run to catalog several large Java applications from different domains. From the results, I show the types of classes developed in these applications, as well as which groups of classes are most commonly developed, which groups of classes are most common within a given domain, and what degree of overlap exists between classifications in different applications and domains. Finally, I draw conclusions about the types of classes being written, and discuss how this work can be utilized to enhance implementation-based testing of Java applications.
