Abstract: To improve the decoupling control performance for multivariable system, a online decoupling control method of neural network based on reference model is proposed. Neural network is constructed to implement feedforward decoupling. Estimate the effect of coupled function on controlled value according to the outputs of reference model and controlled system. Then a learning algorithm of neural network weights parameter is designed and used to adjust the parameters of neural network on-line. The close loop controllers for the decoupling system are designed and the good control performance is achieved. The simulation experiments show the proposed decoupling control method is simple and effective.