Viscosity index (VI) is one of the most important properties that determine the quality grade of lube base oil and it strongly depends on the molecular structure of the hydrocarbon constitute. To produce group III base oil, which is the most valuable base oil type, through lube hydroprocessing, the VI of the final product should be over 120. However, rigorous studies and analyses on the compositional change in lube hydroprocessing and a practical correlation model for VI with respect to the chemical compositions are still lacking. In this study, pilot tests for lube hydroprocessing, which is composed of hydrotreating/cracking followed by hydroisomerization, are implemented with three different types of feedstocks (paraffinic, intermediate, and naphthenic vacuum gas oils) and under different reaction conditions (catalyst and temperature). The fractional changes of the hydrocarbon components by the lube hydroprocessing reactions are quantitatively confirmed using a gas chromatography with mass selective detector. The resulting VI changes of the intermediate and final products are analyzed, and consequently the required reaction condition to manufacture the final lube base oil having a desired VI value is studied for each feedstock type. Finally, on the basis of the experimental data, two regression models for VI are developed: the one is purely data based using a stepwise regression method and the other one considers the physical meaning of the regressed component using the constrained linear regression.