First, you should define what you mean by structural superimposition, especially if you think of superposing distant proteins...

If you proteins are relatives (same fold, or better, same family), then I would suggest refering to the backbone or C-alpha, but pay attention to variable regions. For instance you should first align conserved / structured regions, then try to add external regions (loops) to update you first superposition.

You can achieve this easily using profit, for instance (http://www.bioinf.org.uk/software/), or any distance-based matrix optimization technique (bioperl, biopython, ...).

If you search for a less related set of proteins, then the discussion becomes more opened, but you can have a look at our tool, MulPBA which proved to be more accurate than many existing tools: http://www.dsimb.inserm.fr/dsimb_tools/mulpba/. This tool will allow a more complex superposition of your proteins based on a structural alphabet technique, and should provide you a better superposition than your initial question suggests.

Just to complete the first answer and just in case you are calculating a set of NMR structures... You should give the RMS Capha and all atom, as they give information on the convergence of your calculation. You can give it for all the residues and by subset for sheets and helices.

@ Stéphane the tools you have mentioned work well for structures deposited in PDB. However, i am looking for structural superimposition using homology models. For this purpose i have been using SPDBV But was confused whether to consider Calpha RMS values or backbone as both the values differ

Although the absolute values of theRMSD values maybe different for the Calpha and the backbone atoms, the relative profiles might be very similar, ie. the two RMSD profiles will probably have a high correlation coefficient. If that is the case, then it does not matter if you take one or the other.

Ok .. Thank you