We investigate the capability of global leaf area index (LAI) retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) to be assimilated within a process-oriented biosphere model in order to constrain the distribution of carbon fluxes. This is achieved by implementing a sequential data assimilation procedure within a Dynamic Global Vegetation Model. Assimilating two years (2000–2001) of satellite LAI retrievals advances the onset and the end of the growing season at high northern latitudes, by 20 and 40 days respectively. This reduces the growing season length and leads to lower estimates of global annual gross primary productivity (GPP) and net primary productivity (NPP), respectively by 5 and 3% with large variations from one vegetation biome to another. In situ measurements of monthly GPP from eddy flux towers provide an independent check on the performance of the assimilation procedure, resulting in an improvement of 25% in terms of root mean square error between modelled and observed GPP over the model grid points where the flux towers are located.