The authors describe experiments with a knowledge system that finds sources of research funding based on semantic matches between research proposals and the interests of funding agencies. The GRANT system relies on domain-specific knowledge about semantic matching and a domain-independent partial matching algorithm to search for funding agencies in a semantic network. The semantic matching algorithm implements a model of uncertainty in classification problem solving. The authors analyze cases of poor performance that illustrate how GRANT-like systems are build and refined. They summarize an algorithm for learning domain-specific semantic matching knowledge.