Many decisions can be represented as bipolar, qualitative sets of arguments: Arguments can be pros or cons, and ranked according to their importance, but not numerically evaluated. The problem is then to compare these qualitative, bipolar sets. In this paper (a collaboration between a computer scientist and a psychologist), seven procedures for such a comparison are empirically evaluated, by matching their predictions to choices made by 62 human participants on a selection of 33 situations. Results favor cardinality-based procedures, and in particular one that allows for the internal cancellation of positive and negative arguments within a decision.