A nitrogen-based, seven-component ecosystem model of Fasham [1993] type has been coupled to a three-dimensional quasi-geostrophic ocean general circulation model and applied to a simulation of the seasonal variability of physical and ecosystem variables in the northwestern Indian Ocean. The comparison of the solution with available data (seasonal changes in surface chlorophyll at certain key positions from the climatological coastal zone color scanner (CZCS) database, satellite-derived annual primary production, and data from cruises) has highlighted many agreements but also some disagreements between the model and observations. In particular, the model produces an entrainment bloom in May--June in the central Arabian Sea for which there is no evidence in the CZCS data and does not reproduce the observed August--September bloom in the northern Arabian Sea. In the latter case an analysis showed that zooplankton grazing control prevents the development of the model bloom. A comparison of our solution with the results obtained by McCreary et al. [1996] using a 2.5-layer physical-ecosystem model showed that, despite numerous differences between the models, the period and location for most blooms in these solutions coincide closely. The model reproduced the differences in primary productivity between the northeast and southwest monsoon periods and showed that the model phytoplankton in the main upwelling areas were only seriously nutrient limited during the spring intermonsoon period. For large parts of the year the model predicts that phytoplankton production is closely coupled to zooplankton grazing with blooms only occurring when there are rapid changes in phytoplankton growth rate due either to the entrainment of nitrate into the mixed layer or decreased light limitation when the mixed layer is shallowing thereby allowing the phytoplankton to escape from grazing control. The predicted particle flux from the euphotic zone was of the order of 40% of the primary production and lagged the latter quantity by 8--10 days. This time lag meant that the correlation between daily values of the two quantities was low, and this may partly explain the difficulties of finding good correlations between observations of these two quantities. ¿ 1998 American Geophysical Union