This paper proposed a moving horizon approach to predict sulfur concentrate grade in a sulfur flotation process. In this approach, the grade prediction problem is formulated as a moving horizon estimation problem. To build the nominal model for the estimation, the kinetic of sulfur flotation process are firstly studied. The unknown variable in the kinetic model, i.e., the flux of overflow, is obtained using machine vision technology. Moreover, to account for the multi-modal characteristic of sulfur flotation process, the kinetic model parameters are identified for different working conditions. To eliminate the effect of , a robust parameter identification approach is adopted. Finally, the kinetic model is embedded in the moving horizon estimation framework, which reconstructs the concentrate grade using from online measured process variables. Experimental results demonstrate the feasibility and efficient of the proposed method.INDEX TERMS moving horizon estimation, kinetic modeling, parameter identification, mineral processing