A basic question for research in model-based, qualitative reasoning is how can we predict or analyze the behavior of complex systems without resorting to completely quantitative models. One difficulty that arises is the ambiguity of results due to conflicting, qualitative indications. Argumentation has long been recognized as a means for resolving issues of belief in situations characterized by incomplete, uncertain, inconsistent, and imprecise knowledge. We explore the application of a model of dialectical argumentation to the domain of qualitative reasoning. Models take the form of qualitative networks, with variables connected by strict or default, positive and negative arcs. A notion of defeat between qualitative arguments, which are represented as paths in a network, is defined. Burden of proof is specified as a flexible means of allocating risk, determining relevant argument moves and eventual outcomes. Examples are presented that illustrate semantics of the approach.