This study deals with the design of an inferential estimator to predict matte grade in a copper smelter. A novel mathematical model for the dynamics of the smelter is presented and used as the basis of the on-line estimators. Two different types of estimators are proposed : an estimator with fixed parameters and an estimator with parameters which adapt in time based on infrequent process measurements. Several configurations of the adaptive and fixed estimators, based on different combinations of parameters and measurements, have been designed and compared to plant data. Conclusions are drawn regarding the most accurate configuration.