<p>Snow cover properties have a large impact on the partitioning of surface energy fluxes and thereby on near-surface weather parameters. Snow schemes of intermediate complexity have been widely used for hydrological and climate studies, whereas their impact on typical weather forecast time-scales has received less attention. A new multi-layer snow scheme is implemented in the ECMWF Integrated Forecasting System (IFS) and its impact on snow and 2-metre temperature forecasts is evaluated. The new snow scheme is evaluated offline at well instrumented field sites and compared to the current single-layer scheme. The new scheme largely improves the representation of snow depth for most of the sites considered, reducing the root-mean-square-error averaged over all sites by more than 30%. The improvements are due to a better description of snow density in thick and cold snowpacks, but also due to an improved representation of sporadic melting episodes thanks to the inclusion of a thin top snow layer with a low thermal inertia. The evaluation of coupled 10-day weather forecasts shows an improved representation of snow depth at all lead times, demonstrating a positive impact at the global scale. Regarding the impact on weather parameters, the use of the multi-layer snow scheme improves the simulated daily minimum 2-metre temperature, by decreasing the positive bias and improving the amplitude of the diurnal cycle over snow-covered regions. The analysis indicates that a more realistic representation of snow processes is essential to improve the simulation of low temperature extremes at high latitudes, where snow is a key component of the climate system. The work also highlights that other errors in polar regions still need to be addressed, such as cloud radiative properties, despite the improvements in the responsiveness of snow-covered surfaces with respect to the atmospheric forcing.</p>