Wind tunnel experiments and Computational Fluid Dynamics (CFD) are used to analyse the flow conditions in a venturi-shaped roof, with focus on the underpressure in the narrowest roof section (contraction). This underpressure can be used to partly or completely drive the natural ventilation of the building zones. The wind tunnel experiments are performed in an atmospheric boundary layer wind tunnel at scale 1:100. The 3D CFD simulations are performed with steady RANS and the RNG k-e model. The purpose of this study is twofold: (1) to evaluate the accuracy of steady RANS and the RNG k-e model for this application and (2) to assess the magnitude of the underpressures generated with different design configurations of the venturi-shaped roof. The CFD simulations of mean wind speed and surface pressures inside the roof are generally in good agreement (10-20%) with the wind tunnel measurements. The study shows that for the configuration without guiding vanes, large negative pressure coefficients are obtained, down to -1.35, with reference to the freestream wind speed at roof height. The comparison of design configurations with and without guiding vanes shows an – at least at first sight – counter-intuitive result: adding guiding vanes strongly decreases the absolute value of the underpressure. The reason is that the presence of the guiding vanes increases the flow resistance inside the roof and causes more wind to flow over and around the roof, and less wind through it (wind-blocking). As a result, the optimum configuration is the one without guiding vanes.