An expert system has been developed to support blast furnace operations in the integrated iron and steelmaking works. In the present study, a technique to diagnose furnace abnormality, using the artificial neural network, is proposed to overcome problems of the conventional rule-based expert system, such as lack of in-system automatic regulation and the limits of knowledge expression, etc. In order to enhance diagnostic resolution and robustness to counteract various furnace abnormalities, the data acquired by several sensors installed in the blast furnace were pretreated before using them as imput sources for diagnosis. For the diagnosis of abnormal conditions, back-propagation type of artificial neural networks were constructed, which consist of input, hidden and output layers. Diagnostic procedures are structured hierarchically, that is, first the diagnostic network finds the abnormal conditions and then carries out a detailed diagnosis ac-cording to the type of abnormality. In this study, an action guidance is suggested through the analysis of actual operation results according to the type and progress of the abnormal conditions. The expert system was tested using the operation data obtained from the abnormalities in a real blast furnace. Without the expert system, the operator could not recognize the initial irregularity due to the gradual change of furnace conditions. On the other hand the expert system could find the initial furnace abnormality and suggested appropriate counteractions.