Trade costs are a crucial in new economic geography (NEG) models. The unavailability of actual trade costs data requires the approximation of trade costs. Most NEG studies do not deal with the ramifications of the particular trade costs specification used. This paper shows that the specification of trade costs matters. Estimations of a NEG wage equation for a sample of 80 countries show how the relevance of the key NEG variable, market access, depends upon the trade costs specification. Our conclusion is that NEG needs to (re-)examine the sensitivity of its empirical findings to the handling of trade costs.