The main problem of geostatistical simulations is to reproduce the real characteristics of the mineral deposit when there are few conditioning data. In this case, the geological interpretation, based on structures, geophysics and geological maps, plays the major role in terms of resource evaluation.The paper investigates the use of principal component analysis (PCA) to help classify multiple geostatistical categorical simulations considering the similarity among them and the interpreted model, herein referred as training image (TI).An iron deposit with four lithological facies, in a specific horizontal plane, was used as case study for two-dimensional (2D) simulations. Sequential indicator simulations (SIS) was used to obtain ten equally probable 2D images, which were generated and compared against the interpreted model (TI).In this study, 11 random variables are comprised by the vector of ten SIS realisations plus the training image (TI). The similarity between the TI and each SIS realisation is measured using the angle between the TI and SIS vectors projected in the new coordinate systems (PCA factors). It is possible to order the SIS realisations from the highest to the lowest PCA factor taking into account the TI similarity.The visual analysis of the images shows coherence in the PCA results as a method to measure similarity. This methodology can be used as a post-processing tool to rank simulation results and to select the ones which better represent the prior known geology or to compare the quality of different methods of simulation.