A smelter model has been developed as the basis for an inferential matte grade estimator for an industrial copper smelter. This model couples dynamic mass balances on each species with equilibrium relationships for each component to form a system of differential and algebraic equations. An extended Kalman filter (EKF) based on the dynamic model was applied to track variations in feed composition and enhance the model performance for estimating matte grade. For simulated plant data, the extended Kalman filter tracked feed composition disturbances very well. The effect of the frequency of matte grade measurements on the accuracy of the grade estimates was studied. Off gas composition measurements were also used to infer matte grade and improve the estimator performance. When these secondary measurements were used in conjunction with the matte grade measurements in the EKF algorithm, the estimator produced the smallest error between the estimate and the plant.