The generalized maximum likelihood principle and other statistical regression techniques were applied to 21 sets of ternary LLE
VLE and VLLE data for estimating the pair parameters of three local-composition models (NRTL
FHW and UNIQUAC)
and the corresponding fitting accuracies were compared. Calculational results showed that the correlation and prediction of multicomponent fluid phase equilibria may be greatly improved by applying the maximum likelihood principle. Furthermore
the reliability and limitation of the cross-prediction among LLE. VLE and VLLE were investigated by utilizing the pair parameters obtained from binary and ternary data for the three models. The gap and interdependence between LLE and VLE were also discussed.