A five-component, data assimilative marine ecosystem model is developed for the high-nutrient low-chlorophyll region of the central equatorial Pacific (0N, 140W). Identical twin experiments, in which model-generated synthetic 'data' are assimilated into the model, are employed to determine the feasibility of improving simulation skill by assimilating in situ cruise data (plankton, nutrients and primary production) and remotely-sensed ocean color data. Simple data assimilative schemes such as data insertion or nudging may be insufficient for lower trophic level marine ecosystem models, since they require long time-series of daily to weekly plankton and nutrient data as well as adequate knowledge of the governing ecosystem parameters. In contrast, the variational adjoint technique, which minimizes model-data misfits by optimizing tunable ecosystem parameters, holds much promise for assimilating biological data into marine ecosystem models. Using sampling strategies typical of those employed during the U.S. Joint Global Flux Study (JGOFS) equatorial Pacific process study and the remotely-sensed ocean color data available from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS), parameters that characterize processes such as growth, grazing, mortality, and recycling can be estimated. Simulation skill is improved even if synthetic data associated with 40% random noise are assimilated; however, the presence of biases of 10-20% proves to be more detrimental to the assimilation results. Although increasing the length of the assimilated time series improves simulation skill if random errors are present in the data, simulation skill may deteriorate as more biased data are assimilated. As biological data sets, including in situ, satellite and acoustic sources, continue to grow, data assimilative biological-physical models will play an increasingly crucial role in large interdisciplinary oceanographic observational programs.