I am looking for the relationship between one continuous response variable (hormone levels) and 6 independent variables, some continuous and some categorical. My plan is to perform multiple regression with various combinations of independent variables, then use AIC to determine which of these combinations yields the most precise and parsimonious model. However, with 6 variables, this would be a rediculous number of models to compare. So, what is an appropriate way to select variables before I start this process? In the past, I used backwards, forwards and "all subsets" variable selection. This approach seems to be out of vogue; I assume that the # of paramaters used was not taken into account and that is why AIC is preferred?? Thanks for your assistance and please keep your answer as simple as possible, I am a biologist, not a stats guru (obviously).