Advance control of semiautogenous (SAG) grinding has become a major issue for mining companies. SAG mills are very complex systems involving important nonlinearities and unmeasured disturbances which make the process very dificult to control. In the last decades approaches varying from decentralized PID control to multivariable predictive control have been applied to control of SAG mills with variable success; at the same time, new sensors has been developed with the aim to provide real time measurements of important variables currently unmeasured. The recent availability of new measurements, such as ore size distribution, has faced control engineers with the problem of obtaining new models, and designing new control algorithms capable of both, providing mathematical relations describing SAG mills' behavior, and incorporating the new measurements in determining the control law. In this work, real time measurements obtained using a WipFrag R visual sensor are used in order to obtain a qualitative characterization and a dynamic model of SAG mill's performance when faced to disturbances in the ore size distribution. Results obtained show that the two parameters of the Rosin-Rammler distribution, n and xc, delivered by WipFrag R are the best option for characterizing the ore size distribution. Modeling results show that including the information regarding ore size distribution reduces the mean absolute error from 2:91% to 2:33% with respect to the case when only the feed rate and the speed of the mill are used to obtain the model. © 2011 IFAC.