The work of Heric and Brewer (1969 Journal of Chemical Engineering Data, 14, 55–63) which involved testing a number of liquid viscosity correlations resulted in concluding that the McAllister (1960 A.I.Ch.E. Journal, 6, 427–431) model was the most accurate. However, the fact that the McAllister model is correlative in nature severely limits its practicality and usefulness. This is because costly and time-consuming data are required for the determination of the adjustable (or interaction) parameters contained in that model (Asfour, Cooper, Wu & Zahran, 1991, Industrial and Engineering Chemistry Research, 13, 1666–1669). This study reports the development of a generalized expression of the McAllister model for multi-component liquid mixtures, evaluation of the generalized McAllister model parameters, converting the McAllister model into a predictive model, and comparison of the predictive capability of generalized McAllister model with those of the GC-UNIMOD reported by Cao, Knudsen, Fredenslund and Rasmussen (1993a,b Industrial and Engineering Chemistry Research, 32, 2077–2087, 2088–2092) and with the generalized corresponding states principle (GCSP) which was reported by Teja and Rice (1981 Industrial and Engineering Chemistry, Fundamentals, 20, 77–81). The comparison clearly indicated that the generalized McAllister model is consistently far superior to the GC-UNIMOD and the GCSP in predicting the viscosities of ternary, quaternary, and quinary liquid mixtures. This, no doubt, represents a significant accomplishment in the area of predicting the viscometric behaviour of multi-component liquid mixtures.