When laboratory-measured relative permeability curves are upscaled for use in a reservoir simulation model, a different set of pseudos may be generated for each gridblock and each flow direction. In practice, it is not possible to work with so many different sets, and the pseudos must be divided into a small number of groups, such that all the pseudos in a particular group are sufficiently similar. In this paper, statistical techniques for multivariate data analysis are used to perform this grouping, as well as to determine the appropriate number of groups. Various representations of the pseudos prior to grouping are compared. An example is presented where the number of different sets of pseudos was reduced from 614 to 12 without significant loss of accuracy in the simulation results.