Obtaining pure compound parameters and discerning among sets of parameter values that seem to perform equally well are major challenges when fitting equations of state (EoS) of the SAFT family. A new method to improve pure compound parameters estimation for SAFT EoS is proposed and applied to PC-SAFT. The method consists in using information from a sensitivity analysis to define an improved variable-weight cost function and fixing the least sensitive parameter. Using temperature dependent weights and fixing the least sensitive parameter, the model is fitted with a cost function skewed toward the properties and temperature ranges where the EoS is more sensitive. To assess this approach, the performance of parameters fitted with a classical unweighted cost function were compared with that of parameters fitted with the new approach. Results from four 1-alkanols and four n-alkylamines show that parameters fitted with the proposed methodology, when compared with those fitted with the classical method, correlate non-associating parameters with molecular mass better. In addition, they fit and predict mixture properties more accurately. Average binary VLE fitting error in liquid composition was reduced from 7.2% to 6.5% and in vapour composition from 3.6% to 3.1%. In addition, mixture liquid density prediction error was reduced from 0.53% to 0.31%. Finally, ternary VLE average prediction error was reduced 15% for vapour compositions and 43% for temperature. (C) 2016 Elsevier B.V. All rights reserved.