Cascade control is a frequently used control strategy in industrial processes. It can reduce the maximum deviation and the integral error compared with the single loop. Currently, the minimum variance method of parallel cascade control system are developed based on the assumption that all the disturbance are subject to Gaussian distribution. However, in the practical condition, some disturbances do not obey the Gaussian distribution. The minimum entropy index of performance assessment of the parallel cascade control system subjected to non-Gaussian disturbances is proposed. In the stochastic process, the information entropy has more general significance than mean or variance for any random variable. The estimated ARMA model for the parallel cascade control loop based on the minimum entropy instead of the minimum mean squares error has better performance for non-Gaussian disturbances. The result of proposed methods are demonstrated through a simulation example.