Different possiblities

-wrong/incomplete structure model. The average scattering power of the considered site is located outside of the range f(A) occ(A) + f(B) (1-occ(A)). So you may have a contribution from a lighter atom/vacancies or from a havier atom.

-The occupancy is trying to compensate for other systematic errors in your refinement (texture, absorption).

-Typically the occupancies are highly correlated with the atomic displacement parameters. Typically the ADPs are contrained to be equal for two atoms sitting on one site because otherwise the refinement will blow up. But in reality one does not expect that the ADPs are exactly equal! So even if everything else is o.k. do not believe to uncritically in refined occupancies....

By the way:

Your sample does not have an occupancy parameter...

Thanks for your answer Andreas Leineweber sir,

               The structural model is matching with the experimental data. I didn't refined ADP's i have fixed them to its ambient values. whether it is acceptable to refine site occupancy of atoms by keeping all other refined parameter fixed?. or is there any other way to get them?

Why do you conclude,  that the structural model matches the experimental data if your analysis produces negative occupation factors? About which kind of structure you are talking? Could you post the refinement results? 

You can try to vary at least the isotropic thermal factor B before the occupancies. Then you can fix them and vary the occupancies. It is better to proceed in this way because occupancies and thermal parameters are strongly correlated. As already cited, these negative values can suggest problems during refinement or a wrong structural model at least for what concern the location of atoms. Are you sure that the atom that shows a negative occupation is located on the used site? It has a sufficient x-ray scattering power? If it is a light element it is difficult to locate it!!! 

Marcella, as far as I remember f is proportional to F whereas B behaves exponentionally, i.e. its impact increases with 2theta. In so far I don't think they are strongly correlated. My personal impression is that in B everything is hidden that cannot be seriously explained. The result are often non-explainable B values which are simply refined to get a better fit. Is it really a good idea to refine B first? I wouldn't do this if I do not have certain chemical or physical reasons.  

In fact Gert in many cases is so, the B reach non-explainable values. I gave just an hint of a possible attempt, in many cases the occupancies could reach also unreasonable values for problems of the refinement in general, but also of thermal factors.

I only wanted point out at this specific example of B that it does not make any sense to refine parameters if a) it is not clear what the program is doing, and b) if there is no real physical reasonfor this refinement. The used mathematical algorithms are "stupid" and optimize everything only with the goal to generate a better fit. Unfortunately, mathematically there are many local minima suggesting a physically meaningful result, but there is only one (really existing) correct minima which does neither does have to be described by the used parameters  nor by the applied software. In so far, all these nice software packages are extremly helpful, but only then if the user needs everything (!) that has been implemented in the software. Therefore I can only warn to simply use a Rietveld software and change some parameters (whatever they mean) and believe that the smallest R-value (whatever they mean :-) ) describes the best solution. There are numerous factors - inclusive experimental mistakes and errors - which need to take into account. Therefore my question above: could you post some data? 

Sometimes I feel here like a medical who gets a call: Hey doctor, I have some pain. What is it? We should really try to prevent recommendations if we do not see any data or do not clearly understand what  is actually the question. There are many people reading our comments and used them out of context. 

You are right, in many cases false local minima could be reached and it does not sense try to vary all the parameters to minimize the R factors. It is also true that there are standard methods to vary the parameters in rietveld refinement, that should clearly be adapted to a specific case. It would be always better to know more about the materials and if possible to see the fit. With my answer I did not think to raise a so big problem...

Being myself a programer, from biomolecular world though, gives me another insight. Occupancies are as all other parameters in refinement a consequence of fitting. It is a jiggle between B-values and occupancies which yield the final numbers. One can restrain values of one or the other parameter to a target or into an interval. If both parameters  have no restraints many things are possible including physically impossible numbers for occupancy.

Thanks all for your comments,

                                  structural model by which i meant the initial model. My material is a mixed rare earth sesquioxide , which have a cubic structure. the cations can occupy two positions 8b and 24d. I need to know whether the cations are preferentially occupying in any one of these sites . i have tried to do a refinement by fixing the site occupancies to the weight fraction in which i have prepared the sample. Then i refined the site occupancies. Agreement indices and the refined value of structural parameters are attached along with the refined plot.

The fit seems good so the structural model seems correct.  I think that the background is a little bit high, you can try to reduce the number of coefficients that you used. It seems also that a small amount of an amorphous component could be present. The site that has a negative occupation has the x coordinate variable, did you varied it? If no I would leave it to vary as well as the coordinates of oxygen but with a reduced code for the variation. 

Do I see it correctly that you refine the occupancies of two different RE metals, i.e. of two different 4f elements? Then their form factors are very similar (little scattering contrast) and I will not expect a very stable refinement of the occupancies. The values will not be very reliable if my suspicion is true...

The difference plot of the first two strong reflexes shows an asymmetry typically caused by either an asymmetric profile, or a peak shift. The latter could be explained by a combination of lattice parameter shift and zero shift. But then the shift had to be visible at higher angles as well. However, the it doesn't (which could be an compensation of both groups). The differences of the last reflections is however symmetric. My opinion is that with these differences you cannot see what you want to see (as A. Leineweber already assumes). My recommendation: fix first these differences in the profile. If you only have excess or deficiency intensity, i.e. small positive or negative peaks in you difference plot, then it becomes time to think, how these differences can be explained. Look, substitutions only change the atomic scattering factors, and you want find the best fitting, and from this you (or the program) gets the fraction of each element. However, if - as well-spotted by A. Leineweber - the substituting atoms are very close to each other, their difference in f is small as well so that it is very likely that the accuracy (not precision!) of your result is comparatively bad to your expectations.

Nevertheless, it is good to see the pattern. Thanks for posting it! 

Seems like a powder pattern of Gd2O3 with Mo-anode...

If u refine the profiles correctly;

From the point o view of substitution, Gd is RE2, substitute by higher atomic number (RE1), so at 8b-site RE1 can occupied 26.28%, but not in 24d-site where excess electron is to much but Gd alone is less (anyway 5.1% is significant).

BUT u didn't refine the thermal parameter yet. Reduce the temperature factor to increase the intensity, try  ~0.002660 at 24d, ~0.003166 at 8b and keep O at that value.

PLEASE post: profile, background and/or preferred orientation if present.