This paper reports on the diagrammatic reasoning component of RedSoar, a knowledge system for identifying red blood cell antibodies. Expert blood bank technologists make heavy use of information recorded on paper-based diagrams while solving antibody identification problems. During problem solving there is a constant interaction with the diagram, both as a source of data and as a scratch memory for recording intermediate results. Red,Soar (Johnson, et al., 1991) was designed to mimic the experts’ use of these diagrams. RedSoar accesses the external diagrams in ways similar to those used by experts and displays a limited ability to learn how to use the diagrams more quickly. Our work highlights both the utility and the difficulties of using such approaches for building knowledge systems. The analysis of the diagrammatic reasoning and the approach used are potentiaUy applicable to a wide range of domains.