Aiming at expansion valve control loop in refrigeration system, the predictive control based on double neuron is proposed to control the evaporation superheat. The one neuron is used for on- line real- time estimation of objective dynamic characteristics and for reducing differences between objective response and true response and the other is used for on- line real- time control. The control system has been implemented successfully to solve the nonlinear problem and moreover to overcome the dead-time problem. The simulation results show that this method is feasible with good dynamic responses and stronger robustness, and does its good work for electronic expansion valve control.