Abstract: This paper discribes a design method of multivariable self-tuning controllers for a class of multivariable systems which the number of inputs is uneq ual to the number of outputs.In a quadratic cost function,the control signaland the output derivations from the time varying reference signals are penalized.The optimal control strategy is developed by minizing the cos function.Combining the the equation of the optimal control strategy and the optimal predictor of the auxiliary outputs,the equation of the estimation of the parameters and the structure of the self-tuning controller are obtained. The closed-loop system of the self-tuning controller can trace the time varying reference signals and also adapt to the unstable open-loop systems.