Expert system implementation can take numerous forms ranging from traditional declarative rule-based systems with if-then syntax m imperative programming languages that capture expertise in procedural code. The artificial intelligence community generally thinks of expert systems as rules or rule-bases and an inference engine to process the knowledge. The welding advisor developed at Sandia National Labs and described in this paper, deviates from this by codifying expertise using object representation and methods. Objects allow computer scientists to model the world as humans perceive it giving us a reD' natural way to encode expert knowledge. The design of the welding advisor, which generates and evaluates solutions, will be compared and contrasted to a traditional rule-based system.