Argumentation, and in particular computational models of argumentation, has recently become a main topic within artificial intelligence. This is not only because of its crucial importance and wide applications in other fields of science like philosophy, law, logic, and medicine but also because of its connection to other areas of AI, in specific, knowledge representation. Although there exists a wide variety of formalisms of argumentation, one popular, prominent and simple formalism stands out, namely abstract argumentation frameworks (AFs) first introduced by Dung. Intuitively, an AF is a directed graph in which nodes represent arguments and directed links represent conflicts between arguments. The conflicts between the arguments are resolved on the semantical level. Although AFs are very popular tools in argumentation because of their conceptual simplicity, they are not expressive enough to define different kind of relations. ^Several generalizations of AFs exist, in particular, abstract dialectical frameworks (ADFs), a powerful generalization of AFs, are widely studied. ADFs, first defined by Brewka and Woltran, are capable to express arbitrary relations between arguments with no need of defining a new type of relations and by assigning an acceptance condition to each argument in the form of a propositional formula. In the current work we close some gaps in existing research on ADFs. More specifically, we investigate whether some main results carry over from AFs to ADFs. For instance, we reformulate Dungs Fundamental Lemma and we study under which conditions all semantics of an ADF coincide. We also study whether particular properties which are known to hold for certain subclasses can be extended to the world of ADFs by defining related subclasses of ADFs. ^To do so, we introduce several such classes (symmetric ADFs, acyclic ADFs, attack symmetric ADFs, acyclic support symmetric ADFs, complete ADFs) and investigate their properties. A central aspect of our work is comparing the expressivity of subclasses of AFs and ADFs from the perspective of realizability. At the end we introduce an implementation of a generator to produce such subclasses of ADFs. We use this generator in order to evaluate the effect of cycles on the performance of existing solvers for ADFs.