Abstract: In the process of metallurgical and chemical industry production, some control objects have complicated structures. It's difficult to express them with exact mathematical models and to control them with accurate control method. So at practice,this kind of object is usually controlled by the operators who have rich experience. In this paper, we proposed a semantic interval based approach to automatically generate the knowledge base of a fuzzy controller from the numerical data of the human operation for the interested system. The semi-fuzzy sets are used to express the experts' knowledge. The ID³ algorithm is used to generate the linguistic rule set of fuzzy controller. And using the neural network computation as the inference method, we get the exact output value. The results of the electric stove experiment show the applicability of the proposed semi-fuzzy approach for the design of a controller in real life cases.