Considering the large number of putative Ras effector proteins, it is highly desirable to develop computational methods to be able to identify true Ras binding molecules. Based on a limited sequence homology among members of the Ras association (RA) and Ras binding (RB) sub-domain families of the ubiquitin super-family, we have built structural homology models of Ras proteins in complex with different RA and RB domains, using the FOLD-X software. A critical step in our approach is to use different templates of Ras complexes, in order to account for the structural variation among the RA and RB domains. The homology models are validated by predicting the effect of mutating hot spot residues in the interface, and residues important for the specificity of interaction with different Ras proteins. The FOLD-X calculated energies of the best-modelled complexes are in good agreement with previously published experimental data and with new data reported here. Based on these results, we can establish energy thresholds above, or below which, we can predict with 96% confidence that a RA/RB domain will or will not interact with Ras. This study shows the importance of in depth structural analysis, high quality force-fields and modelling for correct prediction. Our work opens the possibility of genome-wide prediction for this protein family and for others, where there is enough structural information.