In this paper an application of multivariable process identification and robust control to an industrial stabiliser is reported. The requirement is to improve the control of both the bottom quality and the top quality of the stabiliser. The methodology of identification and robust control is introduced briefly. A three-input two-output model of the process is identified using experimental data, where upper bounds of modelling errors are also estimated. The nominal model and the bounds are used to design an robust controller. The controller is tested on the process and significant improvement of product quality and production flexibility has been achieved : the standard deviation of the top product is reduced by more than 93% and that of the bottom product is reduced by more than 37%; setpoint changes take less than 40 minutes.