Grazing dynamics are one of the most poorly constrained components of the marine carbon cycle. We use inverse modeling to infer the distribution of community‐integrated zooplankton grazing dynamics based on the ability of different grazing formulations to recreate the satellite‐observed seasonal cycle in phytoplankton biomass after controlling for physical and bottom‐up controls. We find large spatial variability in the optimal community‐integrated half saturation concentration for grazing (K1/2), with lower (higher) values required in more oligotrophic (eutrophic) biomes. This leads to a strong sigmoidal relationship between observed mean‐annual phytoplankton biomass and the optimally inferred grazing parameterization. This relationship can be used to help constrain, validate and/or parameterize next‐generation biogeochemical models.