The model is evaluated internally and externally. The internal self-consistency check is that the model has to satisfy most restraints used to calculate it, especially the stereochemical restraints. If some restraints are grossly violated in all the models it is likely that the alignment in the corresponding region is incorrect. The restraint violations are reported by the ENERGY command and can be found at the end of the log file. External tests include programs such as PROCHECK by Roman Laskowski and Janet Thornton (anonymous FTP at ftp.biochem.ucl.ac.uk) [Laskowski et al., 1993], the EEF1 web server by Themis Lazaridis (http://mingus.sci.ccny.cuny.edu/server/) and various 3D profile tests including the PROFILE3D program written in the David Eisenberg group (e-mail david@uclaue.mbi.ucla.edu) [Lüthy et al., 1992] and PROSAII written in the Manfred Sippl group (anonymous FTP at gundi.came.sbg.ac.at) [Sippl, 1993]. QUANTA and INSIGHTII also offer various test options.
It is useful to compare the models among themselves because those
regions that are most variable are also likely to be most in error.
Another useful comparison is between the representative model and the
templates. Start by comparing the 
traces, then continue with
the backbone comparison and finally include the sidechains.
The aim of evaluations is to determine whether or not the model is acceptable. If it is not acceptable, that is if the current model violates some restraints, fails the profile tests, or simply does not appear satisfactory, these evaluations should help to re-align the target sequence and the templates for the next cycle of modeling. It may also be that the model violates restraints because the optimizer did not find a good optimum of the objective function. In such a case, you could make the optimization more thorough using options to the `model' routine (Section 3.2).