Petroleum crudes of different geographical origin exhibit differences in chemical composition that arise from formation and ripening processes in the crude. Such differences are transmitted to the fractions obtained in the processing of petroleum. The use of unsupervised classification/sorting methods such as principal component analysis (PCA) or cluster analysis to near-infrared (NIR) spectra for bitumens obtained from petroleum crudes of diverse origin has revealed that composition differences among bitumens are clearly reflected in the spectra, which allows them to be distinguished in terms of origin. Accordingly, in this work we developed classification methods based on soft independent modeling of class analogy (SIMCA) and artificial neural networks (ANNs). While the latter were found to accurately predict the origin of the crudes, SIMCA methodology failed in this respect.