Optimum operation of grinding mills are important for the economic recovery of the valuable minerals, for energy efficiency, as well as from the point of view of pollution control. In many mineral beneficiation operations, the economic objectives translate into the maximization of throughput with suitable constraints on the product particle size distribution. While normal PID type controllers at the regulatory level are capable of controlling the process at the desired values of the process variables, a supervisory control system based on a process model will be required for optimizing the operation. In the present paper the ball mill model, together with the hydrocyclone separation system model is used in a simulation study to generate a response surface relating the control variables with plant load variables and manipulated variables. This response surface together with a suitably formulated economic objective function is used for on-line optimization to determine the optimum setpoints for the controlled variables in the supervisory control system. The model based supervision together with the regulatory layer control has been tested using simulation.