An expert system for algorithm design for industrial machine vision is presented. Its knowledge base includes human experts' knowledge about image processing techniques, and it is capable of solving given vision problems. As a problem domain, the authors have chosen vision algorithms for a parts feeder, which determines the attitude of mechanical parts on a conveyor belt and rejects parts with inappropriate attitudes. The expert system for a parts feeder is designed to consists of three components: a feature selection expert, an image processing expert, and a decision tree generator. The knowledge necessary for the design of a vision algorithm for determining the attitudes of parts is discussed. A framework for representing the knowledge needed in order to find solutions to pattern classification problems is established.