The authors describe an approach to building systems for reading from and interpreting forms of known class. This covers forms where the type of field and relation between fields are known. These cases are more difficult than those where the specific form structure is known, but much less difficult than where there is only a presumption that the input is a form. INFORMys is based on the use of Attributed Relational Graphs, where nodes represent lines, logos, instruction fields, and information fields, and the arcs represent relational dependencies. The graphs are used to guide both form registration and layout analysis, in combination with optical character recognition and supported by connectionist methods of locating logos and instruction fields. The paper reports two evaluation tests with different types of invoice, presenting different system challenges and accuracy/efficiency tradeoffs, with good results. This is a useful state-of-the-art paper.