Since the revival of the gravity literature in the early nineties most work has been done with cross-section data. Just a few more recent contributions made use of panel data. So far that research left open the question whether to treat time and country effects as random or fixed. This note sheds some light on the problems that are associated with a random effects approach. Arguments for the superiority of the fixed effects model arc given both along intuitive and econometric lines based on a Hausman test.