The process of model building followed by refinement can seem to repeat
endlessly. Somewhere it must end. A good time to end would be when the
new model, ready for model building, is so similar to the one from the last
round that the new map will show nothing new. The Fobs are always the same
--
the new map will be distinguished by the difference in Fcalc and 
calc
(the
calculated phase). In my experience, I have never found a difference in a
model when the R value between the two sets of Fcalc is less than 3% and
the r.m.s. phase difference is less than 10 degrees. These parameters should
be monitored each round of model building and can be used to judge when
refinement should stop.