This paper presents the design and application of different fractional order model reference adaptive controllers to a grinding mill circuit. The designed controllers correspond to direct approach, indirect dynamic approach and combine approach. The controller parameters are tuned using an offline optimization process based on nonlinear programming. Its performance is compared to the equivalent non fractional adaptive controllers, designed and tuned using the same procedure, and to a previous adaptive controller designed for this system. Controllers are simulated in the presence of hardness and composition variations in the processed ore, showing a robust behavior, being the direct approach the one presenting the best system performance.