Among the abundance of generalizations of abstract argumentation frameworks, the formalism of abstract dialectical frameworks (ADFs) proved to be powerful in modelling various argumentation problems. Implementations of reasoning tasks that come within ADFs struggle with their high computational complexity. Thus methods simplifying the evaluation process are required. One such method is splitting, which was shown to be an effective optimization technique in other nonmonotonic formalisms. We apply this approach to ADFs by providing suitable techniques for directional splitting (allowing links only from the first to the second part of the splitting) under all the standard semantics of ADFs as well as preliminary results on general splitting.