The robust inferential estimation of polymer properties using stacked neural networks is presented. Data for building non-linear models is re-sampled using bootstrap techniques to form a number of sets of training and test data. For each data set, a neural network model is developed which are then aggregated through principal component regression. Model robustness is shown to be significantly improved as a direct consequence of using multiple neural network representations. Confidence bands for the neural network model predictions also result directly from the application of the bootstrap technique. The approach has been successfully applied to the building of software sensors for a batch polymerisation reactor.