The impact of cloud-affected satellite radiances on numerical weather prediction (NWP) accuracy is investigated. The NWP model used is the HIgh Resolution Limited Area Model (HIRLAM). Its four-dimensional variational data assimilation (4D-Var) system was used to assimilate cloud-affected infrared (IR) radiances from the Spinning Enhanced Visible and InfraRed Imager (SEVIRI). Cloud parameters are modelled internally in the observation operator and used in the radiative transfer calculations. The interaction between the cloud parameters and the model control vector variables is incorporated in the adjoint version of the observation operator, which is used to derive cloud-affected Jacobians prior to the inner-loop minimization of the cost function. The developed framework supports an extensive usage of satellite observations with spatial coverage extended into cloudy regions, which therefore provides additional analysis increments and supports a more accurate description of the atmospheric state. In extended assimilation and forecast experiments the total number of assimilated satellite observations could be increased by approximately 10%. This was associated with a clear indication of a positive impact of cloud-affected radiances on the moisture and geopotential height fields of the NWP model analysis and forecast accuracy when used on top of clear-sky radiance observations. This is revealed by reduced analysis errors of the total integrated water vapour and by reduced forecast errors in the mid and upper troposphere.