The paper describes a method for the design of a multivariable self-tuning controller and presents the results of simulations using this controller. The method uses state-space techniques to assign the closed-loop system poles to prespecified locations. The state vector is made up of process input and output values, so that state estimation is unnecessary. Unity closed-loop gain in steady state is ensured by an additional controller term, and offsets are removed by a separate integral action.