I guess you should first ask yourself whether a refinement of H makes sense. Commonly it doesn't (I would roughly estimate in 99% of all cases). Which kind of sample do you have, and please post the diffractogram.

In think the procedure with soft constraints is in principle reasonable. @Gert: the issue is usually not so much getting direction info about the hydrogen from the refined positions, but it is more about getting the integral effect due to the H atoms considered in your structure model. In most cases the info about the conformations from the non-H backbone is so good, that you can well guess where the H should be. I, however, cannot comment on the technique to use strict restraints in Fullprof. I am only confused by the statement "But now I can not go forward" (therefore I answer): Do you have still a bad fit after introducing the H with soft restraints? Then you will not get a better fit with strict restraints, only a more stable one. If you want to increase the stability of the refinement by the strict restraints, then you may alternatively increase the weighting for the soft constraints or waint for somebody who can explain you the strict restraints (constraints).

Thank you very much for your answers. I had a good fit after introducing the H with soft restraints, but the hydrogen bond lenght and angles was not chemically meaningful. Today I tried to use strict restraints, the refinement solutions are attached to in doc file. It seems to be more reasonable. I have one more question, though I not refined the H atom positions the hydrogen bond lenghts and angles have esd's. How can I handle this problem?

Best regards

@ Andreas: I am not sure how much intereference is hidden in the background or some other parameters which are hard to estimate if the software and all procedures are not so clear. In other words, one can calculate many things with programs, but one should  always estimate whether the extracted effects are physically possible (and not only possibly meaningful) only because they increase the "mathematical" fit. This is the major assumption: we define an absolute correlation between a signal and an assumed simulation which cannot be perfect. In the past it has been show not only once that you can get perfect R values, although the structure was totally wrong. This was a major reason to introduce restrains in order to keep the model realistic. Who tells you that we don't do similar mistakes with soft restrains and H? :-) I don't want argue against this assumption. I only want express that it needs more than a few weeks experience in structure refinement to deal with such specific assumptions. Only because something is possible in a software, it does not mean that you can apply it on every system. 

It is like a trailer hitch. If you have your car equipped with one it depends on several factors whether you can use it or better leave it. On your car your will realize very soon whether it is too weak or the trailer too heavy, or the terrain not suitable. If somebody applies a software there is no warning, and actually only the developer perhaps can tell you whether the used code is applicable for your problem. And in many situations even he/she cannot exclude that something happens which he/she did not predicted. 

@Gert: I fully agree, every, especially comfortable feature, will introduce dangers to semi-experts. I mainly wanted to indicate that leaving the H atoms simply away will introduce errors/bias in the the non-H structure, e.g. too low thermal displacement parameters. Here you have the advantage that at least with modest knowledge on organic stereochemistry you can judge whether the outcoming H sites are reasonable or not. And since the form factor of H decreases rapidly  with sinTh/Lam you will have  contributions of H mainly for low-index hkl reflections, where the precise position of the H atoms contributing to the structure factor is not of too big importance. Of course the question remains, whether you do more harm with leaving out the H totally or with introducing H in a not well-defined and inadequate way.

Dear Kabra,

I used the restraint commands from CFML_Restraints.tpcr file, which is automatically prepared in FullProf Program (dconstrn.hlp using Jdi= 3 command). In this file there are distance restraints (DFIX 2.04998 0.00724 C15 H161) and  (AFIX 36.527 0.686 C20 H201 C21) angle restraints, you can choose any restraint and copy it to your *.PCR file. Or you can write the commands in soft restraints format it is the same. I attached useful pdf files and hope it helps you.

Soft distance constraints:

C20 H201 1 0.00000 0.00000 0.00000 0.95246 0.00001

Soft angle constraints:

C14 H141 C13 1 1 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 35.98 0.05