This may be a useful paper to read, although it is a bit dated:
Acta Cryst. (1999). D55, 479-483
How many water molecules can be detected by protein crystallography?
In general, 3 A resolution is beyond where you can reliably add waters (I start
adding them only at 2.7-2.8 A resolution or
Dear Jeorge,
as you add water molecules to a model, the R _and_ Rfree are bound to drop.
It is important to do it the other way round: let coot add waters, then
check the added waters in coot if they make chemically sense, then run
refmac5. Don't trust automatic procedures more than your
Dear Jeorge
are the water molecules without electron density properly coordinated? These
could be ghosts... Additionally, did you solve the structure by molecular
replacement? If so, in your search model, do you see these water molecules at
the same / nearby location? Again, those could be