This report describes the application of a shell expert system from the medical diagnostic system, Neomycin, to Caster, a diagnostic system for malfunctions in industrial sandcasting. This system was developed to test the hypothesis that starting with a well-developed classification procedure and a relational language for stating the classification model eases the development of a program that diagnoses by selection. The Heracles qualitative modeling environment for the Neomycin medical diagnosis system was used as the basis for Caster, and provided a well-defined diagnostic procedure that structured inferences according to a pattern called heuristic classification, and provided a relational language to state knowledge of taxonomies of malfunctions and abnormal state transitions. The inference procedure of heuristic classification supplied with Heracles reduced knowledge acquisition to constructing classification and state-transition qualitative models. It was concluded that, since the inference problem is general, the power of the knowledge lies in the store of previously encountered problem situations and solutions and of heuristic connections between them. Eight figures are provided. (12 references) (EW)