Countries with diversi ed export baskets take advantage of various bene ts, which are said to foster and stabilize economic growth directly and through indirect channels (e.g. reduced income volatility, positive externalities, spillover e ects). This is especially important in the context of developing economies. However, identifying the true determinants of export diversifi cation is di cult as there exists no comprehensive theoretical or empirical framework to capture all potential factors in their entirety. This paper uses Bayesian Model Averaging to uncover the true long-term roots of export diversi cation among 43 potential determinants,and thus 2 potential models. Our results suggest that only four factors are important in predicting export diversi cation levels over the long run: natural resource rents as a percentage of GDP (100 % posterior inclusion probability), primary school enrollment rates (96 %), population size (25 %), and foreign direct investment levels (17 %). Many prominent candidates turn out to be insigni cant in determining diversi cation levels. Neither policy-related variables (e.g. tari s, freedom from trade regulations or democracy) nor macroeconomic factors (such as trade openness, terms of trade or domestic investment levels) nor geographical remoteness (whether the country is an island or landlocked) play a role. Various robustness checks con rm our results.