The operation of grinding circuits are vital to the success of most mineral processing operations in the provision of effective size reduction whilst maximising throughput. However these same circuits, and especially semi-autogenous grinding (SAG) mills, are highly interactive with product re-circulations and long time delays. This makes the optimal control of these difficult by conventional controllers, such as PID, which generally places a heavy reliance upon operations for control. In order to account for the peculiarities (such as ore type changes) and the abnormal situations that continually beset these circuits, it is necessary to adapt linear control techniques to cater for these situations. The Honeywell Hi-Spec Solutions SmartGrindTM controller is a marriage of multivariable predictive control (MPC) with techniques that provide adaptation of the control to optimise the circuit under all conditions. Optimal control is achieved by incorporating pattern recognition techniques to detect and act on discrete events (eg ore type changes) and adaptation of the MPC to account for known process and equipment non-linearities. Alcoa's Wagerup Refinery is the first site worldwide to have this technology applied to the SAG milling circuits and this has resulted in throughput increases in excess of two per cent and significant improvements in circuit operations. This paper describes the implementation of the control to the refinery's Mill 3 bauxite grinding circuit and the capacity/operational improvements achieved thereof.