Global economic models relies extensively on classical economic theories of land use - Ricardian rent theory and von Thünen location's theory - to project future dynamics of land expansion over the world. In spite of their influence, the effective capacity of these theories to explain land allocation at global scale has never been thoroughly measured and their validity is still debated. To help settle the debate, this paper aims at evaluating the classical theories of land use against the available spatial datasets regarding actual land use, land suitability and land accessibility. This work is carried out using specific econometric methods to deal with spatial autocorrelation and heteroscedasticity. The explained variable correspond to land uses (pastures, total cropland, main crop types). The Ricardian theory and the von Thunen theory are tested by using respectively an index of land suitability and an index of land accessibility as explicative variable. In addition, we use different sources of data to take into account the uncertainty surrounding them. Our results may be used to assess the relevance of global scales scenarios of land use produced by economic models.