Based on the principle of decoupling and neuralnetwork, this paper extends the single-loop single neural control system to multivariable case of the temperature-liquid level two-variable interacting control system in the front box of the pressure net of the papermaking machine. By incorporating static feed-forward decoupling compensation, a learningtype decentralize multivariable control system has been proposed. With a parameter tuning algorithm, the nonlinear single neural controller (SNC) in each loop is able to control a changing process by merely observing the process output error in the loop. The only a priori plant information is the process steady state gain, which can be easily obtained from openloop test. Thus, good regulating performance is guaranteed in the initial control stage, even the controlled object varies later. Simulation results show that this strategy is effective and practicable.