This paper proposes a method to measure and reject disturbances caused by changes in the ore distribution of the feed to a grinding mill circuit. A computer vision system is presented to measure the fraction of rocks in the feed, using only basic equipment and algorithms. A feedforward controller is designed to reject the measured feed size disturbance within the mill before causing further disturbances downstream. The feedforward controller is implemented in simulation to control a nonlinear mill model, and is compared to conventional feedback control. The allowable plant-model mismatch and practicality of feedforward control for industrial applications are investigated. It was found that feedforward control has a noticeable advantage over feedback control if well tuned. The most significant disturbance is the efficiency of the mill to break ore into fine material. Since this is not currently measured in real-time, it limits the performance of a feedforward controller in practice.