Exploring the cause of long-term changes in forest productivity is a difficult problem, especially for intensively managed plantations. Along with environmental changes (CO2 increase, N atmospheric input, climate change), the average genetics of stands evolve between successive generations, and silvicultural practices change rapidly (fertilisation, weed management, plantation techniques, early thinnings). This makes it very difficult to isolate the specific contribution of the environment to the shift in productivity which has been commonly observed in plantations (eg. of maritime pine in SW France). One approach is to improve the knowledge of basic ecophysiological processes and to explore the impact of global change through modelling and simulations. However, some of these processes are still poorly described (eg. carbon allocation), and may be critical for the robustness of model outputs. Furthermore, the heavy experimental approaches needed to calibrate ecophysiological models are generally restricted to a narrow range of ecological situations. For these reasons, semi-empirical models of stand productivity are necessary, provided that they are built on sound ecophysiological concepts, formulated in terms of easily measured variables (such as height and diameter), and fitted on large-scale networks of permanent plots. The purpose of this paper is to illustrate the potential of such a method to produce longterm records of productivity changes. The basic material is a network of 35 permanent plots in Sessile Oak high forests (in the northern half of France), observed since 1920-30. A model was designed to estimate the contributions of age, site and silviculture to stand basal area increment, and to combine these with periodic fluctuations and a long-term trend (function of date). We observed that the history and magnitude of productivity changes were different between locations, ranging from NW to NE France. In 2 out of the 4 forests analysed, basal area increment increased greatly over the past 60 years (30 to 50% according to the model).