Automatically mining protein pathway information from the vast amount of published literature has been an increasing need from the pharmaceutical industry and biomedical research community. This task has been proved to be a formidable one. Many systems have been implemented, but few are practical. Some are too restricted and some are overly ambitious. 1 This paper presents the PathwayFinder system with two key innovations that give the system simultaneously generalization power and practical capabilities: (a) PathwayFinder is designed with appropriate level of users'involvement on information extraction, based on the authors'belief that totally automatic pathway retrieval is beyond the current technology; (b) A novel multi-agent architecture is built to support the need of user-computer interactions and domain extensions. As a result, PathwayFinder is flexible, easy to use, and extendable to be customized to other domains. We have successfully applied the PathwayFinder syste...