Abstract: An improved data dimension decreasing method is proposed to reduce the overlapped information in multivariable system, since a number of overlapped information will affect greatly the correct pick-up of PCs (Principal Components) in PCA (Principal Components Analysis). The overlapped information can be found using the correlation coefficients between standardized variables, then they are weighted and integrated altogether. An PCA-DRBFN (Principal Components Analysis-Distributed Radial Basis Function Network) based soft-sensing model is thus developed using the improved method. The model has been applied to the pure benzene rectifying process in a steel plant to estimate the dry point of pure benzene. Simulation results showed that the proposed method and developed model are both favorably versatile.