Abstract: A new method of self-lcarning optimal control bascd on neural networks is proposed for a formaldehyde process.The convergence of the optimal control algorithm is proved.Optimum conditions are obtained with constrains on quality and reaction rate of production employing the yield as the objective function,It is shown that higher quality of he product can be guaranteed,raw material consumption reduced and the life of catalysts prolonged by this operating condition compared with the practical condition of the plant.