With its added features in the measurement of confidence, and lower demands on the training parameters, the Gaussian process regression (GPR) model has been shown to be a powerful dynamic process model for nonlinear dynamic systems. The authors have developed a recursive GPR model which effectively tracks process dynamics in both sample-wise and block-wise manners. By incorporating an appropriate bias correction technique, the recursive GPR model was further developed to provide better match between the predicted and measured process parameter values. Then the adaptation strategy based on autoregression with exogenous inputs structure was incorporated into the recursive GPR model. The simulation results showed that the recursive GPR model effectively tracked the process dynamics of nonlinear dynamic systems and improve the adaptability of dynamic models. The results of application of the recursive GPR algorithms developed to an industrial propylene polymerization process are presented and discussed. (C) 2011 Elsevier B.V. All rights reserved.