This article presents the first real-time implementation of a recently-developed model-based control method on a pilot-scale, multivariable, liquid-level process that exhibits inverse response and whose actuators can be saturated. Issues in real-time implementation of the control method are discussed, as well as a controller tuning procedure. Two model-based controllers are implemented: one is derived using the process outputs; the other is derived using auxiliary outputs. Servo and regulatory responses presented show the differences in the performance of the two controllers. In real time, one of the model-based controllers eliminates inverse response in one controlled output at the expense of yielding a larger inverse response in the other controlled output. This inverse-response elimination is a real-time validation of the linear and nonlinear theoretical results reported in the systems literature.