Align uses the sequence alignment of the entire protein as a guide, it does a good job for identical sequences, but is easily thrown off by sequence changes. For comparing proteins with lower sequence identity, the super and cealign commands perform better.

However, either algorithm tends to iteratively throw out of the fit residues that do not fit within certain parameters. This is great to get an initial alignment of the structures, However, to reproducibly and quantitatively compare multiple structures, I tend to prefer to define a "structural core" that I align with a pair_fit on an explicitly defined set of Calpha positions. If your ligand binds into a cleft between two domains of your receptor, induced inter-domain motions are most easily seen if you align just one of the two domains, and observe how the relative orientation of the other domain changes as a function of ligand binding.