The general trend in ecosystem modelling is to improve the spatial resolution by shifting from rough box-models to fine 3D models. Despite the continuous speeding-up of computing, 3D models involving numerous state variables may remain intractable, especially for parameter calibration, when processes with long half-life periods (i.e, from years to decades) are introduced, such as the behaviour of organic matter in sediment and population dynamics of benthic species. In these cases, a first approach can be provided by fast-running box-models, if they take into account the most crucial hydrodynamic properties of the system. In a macrotidal shelf sea such as the English Channel, the long-term horizontal transport can be summarized by the tidal residual circulation, and the vertical stratification can be sketched by a two- or three-layered integral model. This paper compares the results obtained in the English Channel area by the same biogeochemical equations of pelagic primary production, coupled to 1) a two-layered box-model 2) a three-layered box-model (i.e., with an intermediate cline layer between surface and bottom ones) and 3) a fine-gridded 3D model. Comparison is focused firstly on thermal stratification and summer dinoflagellate blooms in the north-western Channel and secondly on the haline stratification and the sequence of blooms obtained in the eutrophicated Seine river plume. Comparison shows that box-models act as low-pass filters which reproduce correctly the weekly mean time-course, but greatly reduce the variance locally observed in a tide-oscillating plume region. As far as global characteristics are concerned, such as the annual primary production, or the percentage of variation in annual production after reducing the nutrient loadings, the box and 3D models gave very similar results. This conclusion reinforces the usefulness of using box-models as a first approach in long-term processes, for which a long transient phase is expected before reaching the annual periodic solution. (c) 2006 Elsevier B.V. All rights reserved.