Abstract: The paper proposes an approach that uses soft-sensing method to set up the neural network models of the complex industrial process--nickel flash smelting furnace. They are technological index quality models and yield model for the furnace. This opens up a new application field of neural network modeling. The paper also gives a study of steady-state optimizing control for the furnace. The results show that the modeling and optimization provide better effect in saving energy consumption. Finally ,the paper suggests how to implement on-line steady-state optimizing control to the furnace in the future.