This paper builds upon a cognitive model and a study of how people find faults in a simple device (Ritter & Bibby, 2008). This existing learning model, called Diag, was implemented in Soar and is based on the idea that learning consists of procedural, declarative, and episodic learning. Because Diag is implemented in a version of Soar that is no longer supported, an implementation in an up-to-date cognitive architecture is necessary to understand Soar and to model further aspects of this task. We maintained Diag’s basic structure while reimplementing it in a high-level behavior representation language, Herbal, that generates Soar models and can generate different variants quickly. This newly implemented model, called Diag-H, was validated by comparing its predictions to Diag and the existing data. As a result, it could be shown that Diag-H creates nearly the same results as Diag but also incorporates the advantages of Herbal.