A radiologist often needs to reference relevant prior images of the same patient for confirmation or comparison purposes. To effectively support such needs, we developed a knowledge-based patient image pre-fetching system, addressing several challenging requirements that included representation and learning of image reference heuristics and management of data-intensive knowledge inferencing. The system demands an extensible and maintainable architecture design that is capable of effectively adapting to a dynamic environment characterized by heterogeneous and autonomous data-source systems. We developed a synthesized object-oriented entity-relationship model that is appropriate for representing radiologists' prior image reference heuristics. We detail the system architecture and design of the image pre-fetching system. Our design is based on a client-mediator-server framework that is capable of coping with a dynamic environment. To adapt to changes in prior image reference heuristics, ID3-based multi-decision-tree induction and CN2-based multi-decision induction learning techniques were developed and evaluated. We examined effects of the pre-fetching system on radiologists' examination readings. Preliminary results show that the knowledge-based patient image pre-fetching system more accurately supports patient prior image reference needs than the current practice adopted at the study site and that radiologists may become more efficient, consultatively effective and better satisfied when supported by the pre-fetching system than when relying on the study site's existing pre-fetching practice.