Abstract: A Method for predicting the properties of a complex hydrocarbon mixture which comprises selecting one or more known chemical, perceptual, physical or performance properties of the complex mixture and creating a training set from reference samples which contain characteristic molecular species present in the mixture. The reference samples are subjected to GC/MS analysis wherein the often collinear data generated are treated by multivariate correlation methods. The training set produces coefficients which are multiplied by the matrix generated from a GC/MS analysis of an unknown mixture to produce a predicted value for the chemical, performance or physical property or groups of properties selected.