We investigate how data assimilation and post-processing contribute, either separately or together, to the skill of a hydrological ensemble forecasting system. Based on a large catchment set, we compare four forecasting options: without data assimilation and post-processing, without data assimilation but with post-processing, with data assimilation but without post-processing, and with both data assimilation and post-processing. Our results clearly indicate that both strategies have complementary effects. Data assimilation has mainly a very positive effect on forecast accuracy. Its impact however decreases with increasing lead time. Post-processing, by accounting specifically for hydrological uncertainty, has a very positive and longer lasting effect on forecast reliability. As a consequence, the use of both techniques is recommended in hydrological ensemble forecasting.