In Quantitative Structure-Activity Relationship (QSAR) studies, selection of an appropriate method to assign the partial charge is crucial to derive a reliable model. The prediction accuracy of 3D-QSAR models depends mainly on the method by which the partial charges were calculated. Therefore, we are interested in examining the effects of empirical and semi-empirical partial charges on 3D-QSAR methods, Comparative Molecular Field Analysis (CoMFA) and Comparative Molecular Similarity Indices Analysis (CoMSIA). The feasibility and performance of these charges have been tested on Janus Kinase 2 inhibitors. Both empirical (Gasteiger-Huckel, Del-Re, Gasteiger Marsili, Huckel, Merck Molecular Force Field 94 (MMFF94), Pullman) and semiempirical (Austin Model 1 (AM1), Parameterized Model 3 (PM3), Recife Model 1 (RM1), and Modified Neglect of Diatomic Overlap (MNDO)) charges exhibited statistically significant quality. Among them, MMFF94, MNDO and AM1 charges yielded higher cross-validation correlation coefficient (q 2 ) values whereas Del-Re and MMFF94 charges produced topmost predictive ability (r 2 pred ). Overall, MMFF94 in CoMFA and MMFF94 and MNDO in CoMSIA models was found to be the best charges when both q 2 and r 2 pred values were used as an evaluation criterion. Our results could provide an idea of selection of appropriate charge rather than using default charge to enhance the quality of 3D-QSAR when prediction accuracy and predictive ability were employed.