A major difficulty affecting the control of product quality in industrial polymerization reactors is the lack of suitable on-line polymer property measurements. In this article a scheme is developed to predict melt index and density in a fluidized-bed ethylene copolymerization reactor. Theoretically-based models are derived to predict quality variables from the available on-line temperature and gas composition measurements. Adjustable parameters in these models are updated on-line using infrequent laboratory measurements and a recursive parameter estimation technique. The application of this methodology is illustrated using operating data from an industrial reactor. It is shown that both melt index and density can be successfully predicted. Knowledge of product property deviations from desired targets is required so that manufacturers can take corrective actions to reduce the quantity of off-grade material made and produce a consistent product.