A hybrid non-linear model predictive controller (HNMPC) is developed for a run-of-mine ore grinding mill circuit. A continuous-time grinding mill circuit model is presented with a hydrocyclone cluster as the primary classifier. The discrete-time component is the switching of hydrocyclones in the hydrocyclone cluster. The resulting model is a hybrid non-linear model with both continuous and discrete dynamics. A simulation of the HNMPC shows the advantages of using the hydrocyclone cluster as an additional manipulated variable. The advantages of the HNMPC is illustrated by comparing its performance to a non-linear MPC where no switching of hydrocyclones is possible. The genetic algorithm based HNMPC showed increased controller stability in its ability to incorporate discrete dynamics into the controller directly. The methods discussed in this paper can be used to incorporate different types of discrete dynamics into advanced grinding mill circuit controllers due to the modular presentation of the model and HNMPC controller design.