Hi All, the difference is most likely because of H atoms, indeed (deferent "ideal" values used by phenix.reduce and Monomer Library that is used in phenix.refine). In the next version of phenix.refine the geometry of H atoms will always be regularized (unless free refinement is requested at ultra-high resolution), so it will be always consistent with the Monomer Library definitions. Also, I think H atoms should be excluded from geometry statistics calculations if riding model is used. I will add this as well. If all you need is to regularize the geometry, I would use more specific tool for this: % phenix.pdbtools model.pdb --geometry-regularization You can do it with phenix.refine as well (using wxc_scale=0 and strategy=individual_sites), but don't expect the same outcomes since the default values of number of iterations and number of macro-cycles are different. Also, doing it through phenix.refine will be slower. Pavel.