Abstract: Almost all control loops in industry are under the influence of time-variant disturbances. Performance assessment based on a single model of disturbance is not sufficient for such systems. To solve this, we propose performance assessment of control systems based on a multiple model mixing method with generalized minimum variance. A trade-off is made between the manipulated variable variance and the controller output variance, and the strategy of judgement-weighting is proposed. First, the controller with the minimum generalized variance is chosen and the judgment made as to whether it is different from the previous chosen controller in each period. Then, based on the chosen controllers, a benchmark controller is obtained with the multiple models mixing method. Finally, the generalized variance of the control system under the effect of the benchmark controller serves as the standardwith which to assess a system with time-variant disturbances. Simulation of an ethylene-cracking furnace demonstrates that the proposed method can be applied to control systems with time-variant disturbances.