Agent-based systems are increasingly used to support business-to-business (B2B) electronic commerce and other Internet-based transactions. The design complexity resulting from the multiple interconnected systems in these domains has to be managed in order to reduce costs and time to market. This paper introduces the Role-Algebraic Multi-AgentSystem Design (RAMASD) approach. RAMASD utilizes role models as reusable system-building blocks and a role algebra to capture the basic relations of roles. A two-sorted algebra is used to define the role algebra's semantics. RAMASD reduces the complexity of designing agent-based B2B e-commerce systems by enabling designers to work at a high level of abstraction and by automatically allocating roles to agents according to applicable role models and design constraints. A case study concerning a B2B electronic market for the automotive industry demonstrates the applicability of RAMASD. The advantages an d disadvantages of the proposed approach are discussed, and comparisons with relevantwork are made. Copyright © 2002 M.E. Sharpe Inc. All rights reserved.