Abstract: Self-tuning regulator for multivariable systems with unknown structure and its application to the control of a group of double-input-double-output electrically heated furnaces are discussed in this paper. This includes: determining model structure of the controlled process, giving a succinct algorithm for multivariable minimum variance control strategy, and using exponential-weighted recursive least squares method for on-line identification of controller's parameters. The results indicates the performance of this regulator is better than PID and LQG regulators.