FREEFLAG (the program which is used to generate the test set) 
description says when describing the keyword SEED:

By default, for a given job on a given machine, the random number 
generator produces the same list of "random" free-R flags each time the 
job is run. Since you would generally only produce one list of free-R 
flags for each project, this is not usually a problem. However, if you 
specify the keyword SEED, then the random number generator is seeded 
with the current time, and will produce a different list of free-R flags 
each time the job is run.

So what you see is normal.

Zheng Zhou wrote:
> Sorry that I didn't explain the situation clearly. I used only one 
> output.sca file from HLK2000. I ran the scalepack2mtz (in CCP4i, data 
> reduction, import merged data) several times, on both linux Fedora and 
> window XP, the same computer though.
>
> For the next step refinement, I mean add H2O, ion, ligands, double 
> check certain sidechains.... In those refmac runs, I always use the 
> same mtz file, which I generated from the beginning. I am as surprised 
> as you are the Rfree flags are identical from those different "first" 
> runs.
>
> Thanks
>
> On Jan 23, 2008 9:50 AM, Zheng Zhou <[log in to unmask] 
> <mailto:[log in to unmask]>> wrote:
>
>     Hi, All
>
>     Could any tell me how CCP4 handle free R flag? I know It is
>     important to select the same FreeR reflections if I move to next
>     step of refinement. But everytime I start from fresh, the freeR
>     Flag remains unchanged. The Rwork and  Rfree of my models are fine
>     ( 20.7% and 22.9%). I thought Rfree was randomly generated. I
>     asked more experienced people in the lab and neighbor labs, and
>     didn't get a straight answer. Did I do something wrong?
>
>     Thanks and sorry to bother others.
>
>     Zheng (Joe)
>
> I cant understand this.
> Do you mean you have multiple data sets and they all generate the same
> FreeRs? If you have EXACTLY the same reflection list, and generate FreeR
> flags on the same machine I guess they would be the same.. but it would
> be surprising
>
> Or if you are using the same file which already has FreeR flags then
> they wont change of course - the default is always to use the
> reflections flagged with FreeR = 0.  (Thats as it should be..)
>
> By the by - your agreement between R and FreeR seems unusually low
> unless you have very high resolution data or a great deal of
> non-crystallographic symmetry.
> Eleanor
>
> Zheng  - I am not sure what you are doing, but as long as you work
> with one dataset only, you generate the free reflections at the very
> beginning when you import your data into CCP4. Then use the resulting
> mtz file for all following refinement steps. There are situations
> where one will have to deviate from this scheme, but those are rare.
> If you feel you have such a case, then tell us more about it. Hope
> that helps. Best - MM
>
> Joe,
>
>
> can you explain a bit clearer your problem?  what do you mean the 
> 'next step of refinement'?  are you just doing another round of 
> refinement?  what program are you using, REFMAC5?  also, R/Rfree of 
> 20.7/22.9 is pretty good, depending on resolution.
>
> yes, you are right, Rfree flags are generated randomly and can be 
> anywhere from 5-10% of your reflections (your choice).  then, once the 
> Rfree flagged reflections are selected, they will not change (nor be 
> refined) throughout the rest of your structure determination.  they 
> should have their own column in your *.mtz file called RfreeFlag or 
> something like this (in CCP4 that is).  other programs handle this 
> relatively the same, but may use various naming conventions.  one 
> caveat is that CCP4 flags reflections using 0 by default (meaning 5% 
> of your reflections are flagged with 0 and are the Rfree set).  Other 
> programs such as CNS and PHENIX use 1 by default, so be careful when 
> switching back and forth.  You can tell either program which to use 
> for Rfree set, but needs to be set, since it is not default.
>
> again, not exactly sure what the problem is, but i hope some of this 
> helps.  feel free to email with further questions if needed.  best of 
> luck!
>
>
>
> cheers,
> nick
>

-- 
Edwin Pozharski, PhD, Assistant Professor
University of Maryland, Baltimore
----------------------------------------------
When the Way is forgotten duty and justice appear;
Then knowledge and wisdom are born along with hypocrisy.
When harmonious relationships dissolve then respect and devotion arise;
When a nation falls to chaos then loyalty and patriotism are born.
------------------------------   / Lao Tse /