In this paper, I improved the look-up table of hot coil temperature control system using neural networks, and made the performance of the control system better. Because neural networks can approximate any function using I/O data only, an unknown system can be identified easily. many skills in iron manufacture are based on 'trial and error' and experience of experts. Neural networks are proper things in this field. In the hot coil temperature control system, the look-up table of heat transfer coefficients is the most important thing for the performance. Quantization errors and missing variables are reasons of the low performance. Replacing the look-up table by neural networks trained with good performance data, the predescribed defects were overcome. Neural networks showed superior performance to the look-up table in computer simulations.