Abstract— The design of multi-agent systems is different from similar problems because the concept of agent involves the notions of autonomy and intelligence. As a consequence agent- based software engineering approaches must learn from classical design approaches but should go further introducing an explicit representation of the previous cited notions as well as of ontology, communications, mobility and other agents related issues. Even if these arguments are more than sufficient to justify the study of specific approaches we also think that today a successful design methodology should include some other strategic factors: the use of a standard and well known design language (like UML), the support of a specific CASE tool to simplify the work of the designer and the attention for the automatic production of large parts of code. We propose PASSI (Process for Agent Societies Specification and Implementation) as a solution to the above arguments. It comprehends the construction of five models (System Requirements, Agent Society, Agent Implementation, Code Model and Deployment Model) which include several distinct phases. We also illustrates the contribute of an add-in that we have produced for a commercial UML-based CASE tool in order to have a dedicated design environment that proves more productive of the general purpose ones. Index Terms— CASE tools, Design Methodology, Multi-Agent Systems, Software Engineering. I. INTRODUCTION EVERAL works can be found in literature about the design and representation of multi-agents systems [21][22][23][24][25][26]. Some approaches propose representations involving abstractions of social phenomena and knowledge [21][25][26] obtaining an expressive representation of these aspects but maintaining a distance from the implementation level that introduces a difficulty in the production of the final code solution. On the contrary some others maintain an high level of attention both for analysis steps and implementation issues but are less interested in the representation of the social aspects of the MAS [22][23][24]. Massimo Cossentino is with the CERE/CNR (Centro di Studi sulle Reti di Elaboratori-Consiglio Nazionale delle Ricerche), V.le delle Scienze c/o Centro Univ. Calcolo, 90128 Palermo Italy. (phone: +39-091.6566274; fax: +39.0916529124; e-mail: cossentino@cere.pa.cnr.it). Colin Potts is with the College of Computing of the Georgia Institute of Technology, Atlanta (GA) 30332-0280 USA (e-mail: potts@cc.gatech.edu). We think that multi-agent systems (MAS) differ from non- agent based systems because agents are intended to be autonomous units of intelligent functionality. As a consequence, agent-based software engineering methods must complement standard design activities and representations with models of the agent society. We give also a great importance for the success of a design methodology to some strategic factors: the use of a standard and well known design language (like UML), the support of a specific CASE tool to simplify the work of the designer and the attention for the automatic production of large parts of code (in order to increase the productivity and reduce the number of human errors). We propose PASSI (Process for Agent Societies Specification and Implementation) as a solution to the above arguments. This methodology is the result of a long period of study [1][2][3] and experimentation mainly in robotics [4][5]. It is composed of five models (System Requirements, Agent Society, Agent Implementation, Code Model and Deployment Model) which include several distinct phases. We also produced an add-in for a diffused commercial UML-based CASE tool (Rational Rose) in order to have a dedicated design environment that proves more productive of the general purpose ones. The code production phase is also strongly supported by the automatic generation of a great amount of code. This is possible thanks to the simplicity of the structure of the FIPA architecture [27] that we assume as a reference for our approach, to the use of an XML content language for the messages between the agents (so that the content of each message can be straightforwardly derived from the design) and because of a library of reusable patterns of code. The following sections are organized as follows: in section two we provide a quick overview of the PASSI methodology, in section three we discuss the different phases of PASSI illustrating the contribute of the CASE tool add-in using a simple example coming from robotics, in section four we quickly review the main themes of the paper and address some of the future work issues. II. AN OVERALL VIEW OF THE DESIGN METHODOLOGY Several design methodologies have been proposed in literature (Gaia [26], MASE [9], CASSIOPEIA [28],…) but they didn’t satisfy some of the needs coming from our experiments with multi-agents systems applied to robotics. Several of the previous methodologies are lacking of A CASE tool supported methodology for the design of multi-agent systems Massimo Cossentino, Colin Potts S