We apply graphical modelling theory to identify scal policy shocks in SVAR models of the US economy. Unlike other econometric ap- proaches which achieve identi cation by relying on potentially con- tentious a priori assumptions graphical modelling is a data based tool. Our results are in line with Keynesian theoretical models, being also quantitatively similar to those obtained in the recent SVAR litera- ture à la Blanchard and Perotti (2002), and contrast with neoclassical real business cycle predictions. Stability checks con rm that our nd- ings are not driven by sample selection.