Thanks for the answer and sparing your time to my question. I actually wonder that is it equavalent, to use regression with dummy variables and unviate GLM procedure. For example, the t-tests of coefficients of the dummies gives the same result with the GLM procedure. If I find that a coefficient related to a dummy variable is different from "0", does always GLM procedure results there is significiant difference between means of groups? Thanks a lot.