This paper discusses some soft sensing techniques used for estimating product quality of hydrocracking fractionator. These techniques rely on the statistical analysis, neural networks and the combined method of them. The soft sensing models, which are based on historical process data, were built to predict the jet fuel endpoint online. It also makes a comparison among the results based on partial least squares (PLS), radial base function networks (RBFN) and the combined method of PLS and RBFN (named PLS-RBF algorithm). It shows that the approach based on PLS-RBF algorithm produces more accurate predictions and provides more reliable extrapolation.