We present new empirical and mechanistic models for predicting the export of organic carbon out of the surface ocean by sinking particles. To calibrate these models, we have compiled a synthesis of field observations related to ecosystem size structure, primary production and particle export from around the globe. The empirical model captures 61% of the observed variance in the ratio of particle export to primary production (the pe ratio) using sea-surface temperature and chlorophyll concentrations (or primary productivity) as predictor variables. To describe the mechanisms responsible for pe-ratio variability, we present size-based formulations of phytoplankton grazing and sinking particle export, combining them into an alternative, mechanistic model. The formulation of grazing dynamics, using simple power laws as closure terms for small and large phytoplankton, reproduces 74% of the observed variability in phytoplankton community composition wherein large phytoplankton augment small ones as production increases. The formulation for sinking particle export partitions a temperature-dependent fraction of small and large phytoplankton grazing into sinking detritus. The mechanistic model also captures 61% of the observed variance in pe ratio, with large phytoplankton in high biomass and relatively cold regions leading to more efficient export. In this model, variability in primary productivity results in a biomass-modulated switch between small and large phytoplankton pathways.