An advanced multivariate sequential data assimilation system has been implemented within the framework of the European MFSTEP project to fit a three-dimensional biogeochemical model of the Eastern Mediterranean to satellite chlorophyll data from the Sea-viewing Wide Field-ofview Sensor (SeaWiFS). The physics are described by the Princeton Ocean Model (POM) while the biochemistry of the ecosystem is tackled with the Biogeochemical Flux Model (BFM). The assimilation scheme is based on the Singular Evolutive Extended Kalman (SEEK) filter, in which the error statistics were parameterized by means of a suitable set of Empirical Orthogonal Functions (EOFs). To avoid spurious long-range correlations associated with the limited number of EOFs, the filter covariance matrix was given compact support through a radius of influence around every data point location. Hindcast experiments were performed for one year over 1999 and forced with ECMWF 6 h atmospheric fields. The solution of the assimilation system was evaluated against the assimilated data and the MedAtlas climatology, and by assessing the impact of the assimilation on non-observed biogeochemical processes. It is found that the assimilation of SeaWiFS data improves the overall behavior of the BFM model and efficiently removes long term biases from the model despite some difficulties during the spring bloom period. Results, however, suggest the need of subsurface data to enhance the estimation of the ecosystem variables in the deep layers.