Clinical diagnostic aids are relatively scarce, and are seldom used in routine clinical practice, even though the burden of diagnostic error may have serious adverse consequences. This may be due to difficulties in creating, maintaining and even using such expert systems. The current article describes a novel approach to the problem, where established medical content is used as the knowledge base for a pediatric diagnostic reminder tool called ISABEL. The inference engine utilizes advanced textual pattern-recognition algorithms to extract key concepts from textual description of diagnoses, and generates a list of diagnostic suggestions in response to clinical features entered in free text. Development was an iterative process, relying on sequential evaluation of clinical performance to provide the basis for improvement. The usage of the system over the past 2 years, as well as results of preliminary clinical performance evaluation are presented. These results are encouraging. The ISABEL model may be extended to cover other domains, including adult medicine.