Dear all, sorry to interfere late and also to update the subject (that had nothing to "TLS groups" in the last messages). I would like also to point out an article in which in fact Pavel is one of the co-authors : Lunin et al (2002), Acta Cryst, A58, 270-282 and that is directly relevant to the question of using Fobs=0. Comparison of maximum-likelihood-based and least-squares refinements and a development of the ML-target into a Taylor series show that the ML-refinement has two basic features : 1) it defines in a very special way the weights for some terms 2) for weak reflections, especially when the model is incomplete (that is always the case), it suggests fitting Fcalc to 0 and not at all to the measured Fobs. You may look at the corresponding plots in the article. This means that, complementarily to the French & Wilson study of experimental data, an option to use Fobs=0 maybe not so meaningless in a refinement program especially if by some reasons one wants to use LS-refinement and not ML. Best regards, Sacha PS : My personal request to the phenixbb users - please take care and put a right subject when posting your original message