Hi Nikousaleh, Well that's going to depend on a lot of things. Juat because a model is sterochemically "correct" it is not going to necessarily accurately reflect the true structure of your modelled protein. Statistics are useful to indicate where things have gone wrong, e.g a bad alignment or difficult to model regions, but can also reflect just as much the modelling algorithm/energy minimization/molecular dynamics parameters as anything. You need to make a subjective assessment based on a whole host of criteria: the % identity between template and model; regions with insertions and deletions; "unsatisfied" buried charged side chains; if there is more than one template structure, regions with higher flexibility (e.g loops) versus those with extremely conserved geometry (e.g active site residues), etc. If you're worried about bad contacts you could use Modeller to perform energy minimization of the SCWRL output, but at the end of the day you need to look at the structures, decide what the question is that you are trying to ask of it, and make an assessment yourself of what is the better model. Cheers, James