A large number of high activity catalysts are used to produce different grades of polymers in a continuous olefin polymerization reactor. Often the transition from one grade to another is carried out by changing the reactor operating conditions as well as the catalyst. The major kinetic parameters are seldom known for all the different types of catalysts. The catalyst composition may differ only slightly from each other but enough to result in significant differences in polymer properties. In practice, it is very time consuming and costly to carry out extensive plant tests to identify the kinetics of polymerization for a given catalyst system. In this paper, two different on-line parameter estimation schemes (extended Kalman filtering and nonlinear dynamic parameter estimation) are investigated to estimate key kinetic parameters of transition-metal-catalyzed olefin polymerization. Parameter estimation using an extended Kalman filter is shown to perform robustly even in the presence of substantial measurement noise because of greater flexibility in tuning parameters, while the parameter estimator based on nonlinear programming techniques (nonlinear dynamic parameter estimator) shows a stronger sensitivity to measurement noise due to a lack of sufficient tuning parameters. Both the extended Kalman filter and the nonlinear dynamic parameter estimator give identical results after a sufficient number of measurements have been obtained.