The separation between analysis and design phases has long been advocated in software engineering literature. There has been active interest in the area of agent oriented software engineering but the methodologies developed do not focus on a clear separation between the two phases. Furthermore, existing agent oriented methodologies tend to be tied to a particular design architecture and applicable only for small systems. In this paper, we describe a goal and role based analysis methodology that is both unbiased towards any design architecture and is scalable. The model is derived from improvements to the ROADMAP methodology for agent oriented systems developed at the University of Melbourne. We also present REBEL - CASE tool developed to support the methodology. Furthermore, several examples and experiences with the method are discussed. We conclude by comparing analysis models of other agent oriented methodologies to ours.