In this article the design and implementation of a high-level image diagnosis system are described. A knowledge-based system, OOI, has been developed to incorporate the image processing abilities. It employs a robust control strategy in the object-oriented approach that minimizes the amount of domain-specific control knowledge. Knowledge objects are constructed that embed specialized methods or metaknowledge in image processing. They work independently of each other. The highly modular architecture allows the knowledge engineers to modify the knowledge without worrying about any unexpected side effects. Reasoning in OOI proceeds in both top-down and in bottom-up schemes. The system has been tested in medical image analysis. Cancer cell lesion classification is used as an application domain to illustrate how image diagnosis can be implemented in our system. Results attest to the efficacy of this framework in both image recognition and diagnosis.