How well do biogeochemical data sets serve to decide among models and model parameter values? Data at 21N, 31W from the French JGOFS EUMELI cruises and the SeaWiFS ocean color sensor were used to estimate parameters for three very different models of biological nitrogen flux in a water column. The three models are (1) an NPZD (Nutrients, Phytoplankton, Zooplankton and Detritus) model (Oschlies et al., 2000), (2) a seven-component model with two pools of dissolved organic matter and detritus with different remineralization and sinking rates (Dadou et al., 2001) and (3) a model of nutrients and phytoplankton including aggregates (Kriest and Evans, 1999). Parameters of the three models are estimated using the same sets of data within the same one-dimensional physical framework. A combination of local and nonlocal optimization methods is used. It is not easy to decide among candidate models based on their fit to the data. Parameters that mean the same thing in the three models, like the half-saturation concentration for nitrate uptake, were estimated at not very different values in different models. The model with dissolved organic matter, based on its primary production and sediment flux data time evolutions, seems to exhibit the more reasonable annual behavior. Large seasonal changes in deep nitrate data suggest an unexpected role of lateral advection and may vitiate the 1-D approach even at the EUMELI oligotrophic site. The small number of sediment trap measurements are very powerful for constraining the biological nitrogen. Ocean color data did not add extra constraining power.