Mohammed,

You need to estimate how much of an effect you expect from your intervention.  If in doubt assume a 50% difference in your outcome measure between the intervention and control groups (the "effect size").  Normally you want to be confident that you can detect a difference with 80% power, using a standard measure of significance (normally 0.05 or 1 in 20 but could be lower e.g 0.1 if this is a pilot experiment) between the intervention and control outcomes.  By "guestimating' the  sample, you risk either doing a lot more recruitment and so work than necessary to answer your question, or reducing the power of your test to reject the hypothesis of no difference (accepted when in fact false). Put simply, too small a sample will often not find a differences if one exists; too large and you do a lot of unnecessary work (though the upside is a more precise estimate of any actual difference).  My advice is to use a good sample size calculator.

 Dear Karin Riemann-Lorenz 

Daer Richard Fielding

Thank you for replying me. Your comments were useful.

Best Regards.

M. Malekzadeh