Abstract: Many chemical processes have mechanism complexity and strong nonlinearity. To overcome the shortcomings of traditional modeling methods and improve the prediction accuracy of the soft-sensor model, in this study, a modeling method is proposed on the basis of kernel principal component analysis (KPCA) bagging ensemble neural networks for a chemical process soft sensor. KPCA is first applied to compress the input data of the soft-sensor model, and extracted nonlinear principle components are then used as the input of the model. Second, by bagging the ensemble learning algorithm, several sample subsets are achieved from the original dataset, and these are used to construct multiple back-propagation neural networks. In addition, the numbers of hidden layer units and ensemble submodels are optimally determined using the grid search method. Finally, output fusion of the submodels is realized using ridge regression, and a soft-sensor model is constructed on the basis of KPCA bagging ensemble neural networks. Simulation results for the polypropylene melt index soft sensor indicate that a soft-sensor model has enhanced prediction ability based on the above modeling methods.