A shape learning system is proposed to form a concept of object from the sequences of arbitrary shapes of the world objects, using a cognitive shape representation technique that describes the structure of shape in terms of meaningful parts and their relations. The symbolic description of an input shape is constructed by a cognitive shape decomposition and representation. The input description is compared with a model by traversing the hierarchical objects knowledge base. The learning module creates and refines the concept of each example by generalization and specialization with the result of a matching and the performance standard of teacher. In other words, hierarchical object knowledge base is constructed automatically to recognize fast and categorize an object conceptually. A number of experiments are conducted on different types of shapes. It is shown that the constructed concepts in the hierarchical knowledge base correspond with the concepts formed by human.