This paper investigates the link between trade and convergence in per capita income by applying a threshold methodology to standard growth regressions in order to capture a nonlinear effect of trade on growth. We use 10 trade measures, divided into trade intensity ratios and measure of trade restrictions. For the former group of indicators only, our tests show that linear models have to be rejected in favour of threshold regressions. We have identified three different regimes, characterized by different relationships between growth and its determinants. Countries belonging to the regimes grouping higher-income economies and poorer countries are diverging. The conditional convergence hypothesis is only accepted in the regime of middle-income countries; a process of catching-up can be mainly found in the case of few developing countries. For the vast majority of developing countries, divergence in per capita income seems to be the norm. The correlation between measures of trade restrictions and growth is different across regimes, and the positive effect of a decrease in tariffs on growth depends on the level of development; for the majority of the developing countries included in our sample a decrease in tariffs will have no effect on growth.