Statistical modelling was applied to a large number of historical nutrient data to assess the significance of human perturbations in the Mediterranean Sea. All available phosphate data were grouped into subsets representative of averaged values of the measured vertical profiles in the surface and deep water layers. In contrast to earlier predictions, the statistical analysis of the phosphate concentrations in a deep water layer does not indicate any particular trend in time for the last 30 years. These data sets were then used as an input to an inverse model and a 3D primitive equation model (PEM). The former redistributes the measured concentrations by means of a variational principle and reconstructs average horizontal space distributions of the phosphate data fields as gridded solutions over the whole area. The spatial and temporal distributions thus obtained are visualised graphically and compared with other existing data, providing the first overall view of phosphate in the whole Mediterranean Sea and revealing an increasing oligotrophy towards the eastern basin. The primitive equation model is then used to assess the variability and upwelling fluxes. (C) 2001 Elsevier Science Ltd. All rights reserved.