Dynamic matrix control (DMC) is implemented on a nonlinear, multivariable, experimental pressure tank. Linearized models for the pressure tank are identified using subspace methods, which prove to be suitable over a limited operating range. When the DMC algorithm is implemented, some tuning/design parameters are easy to select but others require trial and error. Here, a novel approach to the selection of these other parameters using response surface methodology is used. Finally, the DMC algorithm is implemented on-line and leads to good control.