Substantial improvements are required in the current suite of numerical models if we are to better understand the present ocean biogeochemical state and predict potential future responses to anthropogenic perturbations. At present, major impediments to marine biogeochemical modeling include the inadequate representation of multi-element cycling and community structure, large-scale physical circulation, mesoscale space and time variability, and mass exchange between the open ocean and the atmosphere, land, and coastal ocean. Marine biogeochemical modeling is inherently data driven, and significant progress on any of these topics will require close collaboration between the observational and modeling communities. Two main thrusts should be to develop improved, mechanistically based parameterizations of specific biogeochemical processes and to test the overall skill of integrated system models through detailed model-data comparison of both the mean state and seasonal to interdecadal variability. ¿ 1999 American Geophysical Union