Flood forecasting systems are today recognized as a key element in natural hazard mitigation. The objective is to exploit the available observed and forecasted meteorological information to foresee river discharges up to several days in advance, using a hydrological model. The quality of the hydrological forecast is directly dependent on the quality of the meteorological input feeding the model. Optimal usage of the meteorological data is therefore essential. Depending on the sources of data considered and the method used to spatialize and combine the data, the interpolated precipitation intensity and spatial distribution may vary considerably. Calibration of the model is an important step in the preparation of the system. The parametrization of the model is adapted to obtain simulated discharges as similar as possible to the observed ones. Finally, by integrating in real time discharge and other observations in the computation scheme with data assimilation methods, the quality of the hydrological forecast can be further enhanced.