Aim: To provide the first global quantification of the slope and shape of the latitudinal gradient in seed mass, and to determine whether global patterns in seed mass are best explained by growth form, vegetation type, seed dispersal syndrome, or net primary productivity (NPP). Location: Global. Methods: We collected seed mass data for 11,481 species × site combinations from around the world. We used regression to describe the latitudinal gradient in seed mass, then applied general linear models to quantify the relative explanatory power of each of the variables hypothesized to underlie the latitudinal gradient in seed size. Results: There is a 320-fold decline in geometric mean seed mass between the equator and 60°. This decline is not linear. At the edge of the tropics, there is a sudden 7-fold drop in mean seed mass. The strongest correlates of the latitudinal gradient in seed mass are plant growth form, and vegetation type, followed by dispersal syndrome and NPP. A model including growth form, vegetation type, dispersal syndrome and NPP explains 51% of the variation in seed mass. Latitude explains just 0.2% of the residual variation from this model. Main conclusions: This is the first demonstration of a major decrease in seed size at the edge of the tropics. This drop in seed mass is most closely correlated with changes in plant growth form and vegetation type. This suggests that the drop in seed mass might be part of a sudden change in plant strategy at the edge of the tropics.