This study applies multilevel modelling to the analysis of trade flows. This methodology can complement the use of the standard gravity equation in addressing the influence of geographical distance on the impact of trade agreements on bilateral trade (the natural-partner hypothesis). The analysis suggests that the simplest, most obvious, and best way to nest country pairs with international trade data is by distance. To this end, a sample of 69 countries with data on domestic flows over the period 1986–2016 was used. This sample accounts for around 85% of world trade data. The estimations suggest that the pro-trade effects of preferential trade agreements decrease with distance to trade partners, while this does not occur with the multilateral trading system. What is more interesting is that the former result only holds for the preferential trade agreements involving goods, which supports the natural-trading partner hypothesis, but not for those that include services. In the latter case, the impact increases with distance. This suggests that deeper preferential trade agreements reduce the cost of distance as a barrier to trade, which probably reflects cultural (regulation) distance.