In this paper we describe PADUA, a protocol designed to enable agents to debate an issue drawing arguments not from a knowledge base of facts, rules and priorities but directly from a dataset of records of instances in the domain. This is particularly suited to applications which have large, possibly noisy, datasets, for which knowledge engineering would be difficult. Direct use of data requires a different style of argument, which has many affinities to case based reasoning. Following motivation and a discussion of the requirement of this form of reasoning, we present the protocol and illustrate its use with a case study. We conclude with a discussion of some significant issues highlighted by our approach.