I develop a multi-country, multi-industry model of trade that features heterogeneous consumers with non-homothetic preferences. I use the model to quantify the measurement errors in the welfare gains estimates caused by the assumption of a representative consumer (ARC). First, I reduce the world level of all trade costs by 15% and find that ARC overestimates (underestimates) the gains of the poor (rich) by up to 5 (11) percentage points. Second, I eliminate import tariffs around the globe and show that the loss of tariff revenues is not negligible for some consumers and that the measurement errors from ARC are between −15 and 4% points