Based on an empirical gravity model of sectoral bilateral trade, we uncover three features of bilateral trade balances. First, the difficulty of gravity models in fitting the observed level of bilateral balances is likely due to the presence of unobservable bilateral trade costs. Second, the model fit improves drastically when we focus on changes over time of the balances. Third, using a log linear approximation we show that changes in bilateral trade balances over the past two decades were driven almost entirely by changes in the same macro factors that determine countries’ aggregate balances – changes in bilateral trade costs, including tariffs, played therefore only a negligible role. This conclusion provides new support for the view that bilateral balances are, for practical purposes, not relevant to the conduct of macroeconomic policy.