Surface-enhanced Raman scattering (SERS) exploits surface plasmons induced by the incident field in metallic nanostructures to significantly increase the Raman intensity. Graphene provides the ideal prototype two-dimensional (2d) test material to investigate SERS. Its Raman spectrum is well-known, graphene samples are entirely reproducible, height controllable down to the atomic scale, and can be made virtually defect-free. We report SERS from graphene, by depositing arrays of Au particles of well-defined dimensions on a graphene/SiO(2) (300 nm)/Si system. We detect significant enhancements at 633 nm. To elucidate the physics of SERS, we develop a quantitative analytical and numerical theory. The 2d nature of graphene allows for a closed-form description of the Raman enhancement, in agreement with experiments. We show that this scales with the nanoparticle cross section, the fourth power of the Mie enhancement, and is inversely proportional to the tenth power of the separation between graphene and the center of the nanoparticle. One important consequence is that metallic nanodisks are an ideal embodiment for SERS in 2d.