A quick question about diffraction data input. I just noticed that phenix.refine uses Imean/SIGImean and converts these by default in my refinements. Is this the preferred way or would it be better to use F/SIGF (after CCP4)?

phenix.refine picks I over F only because that's more likely to be the original measurements. (Internally it always works with F, as you noticed.) I'd expect very similar results starting with either I or the corresponding F, i.e. I don't think it matters. Ralf