This dissertation describes various aspects of improvements made in the representation of clouds in the global forecast model of the European Centre of Medium-Range Weather Fore- casts (ECMWF). Cloud parametrization has long been identified as one of the most crucial and uncertain aspects in General Circulation Models (GCMs) of the atmosphere, which are used for both Numerical Weather Prediction and the simulation of climate. It is therefore important to constantly monitor and improve the performance of cloud parametrizations in those models. The first part of the work describes the implementation of an existing cloud parametrization into ECMWF's forecasting system with special attention to a new treatment of the prognos- tic cloud variables in data assimilation. This is followed by an analysis of the performance of the parametrization during a 15-year long data assimilation experiment carried out in the context of the ECMWF reanalysis project. It is shown that despite an overall good perfor- mance, several weaknesses in the simulation of clouds exist. Subtropical stratocumulus and extratropical cloudiness are underestimated, while the cloud fraction in the trade cumulus areas and in the Intertropical Convergence Zone is overestimated. In the second part of the study detailed revisions of the parametrization of cloud generation by convective and non-convective processes are described. A consistent new description of cloud generation by convection is derived using the mass- ux approach. Furthermore an improved description of the generation of clouds by non-convective processes is introduced. The superiority of the new formulation compared to the existing one is demonstrated and links to other approaches to cloud parametrization are established. The third part of the work studies the role of vertically varying cloud fraction for the descrip- tion of microphysical processes. It is shown that the commonly used approach of representing precipitation in GCMs by means of grid-averaged quantities leads to serious errors in the parametrization of various physical processes such as the evaporation of precipitation, with severe consequences for the model's hydrological cycle. A new parametrization of the eects of vertically-varying cloud fraction based on a separation of cloudy and clear-sky precipita- tion uxes is developed and its performance assessed. It is shown that this parametrization alleviates most of the identied problems and thereby more realistically describes the pre- cipitation physics in the presence of cloud fraction variations. The final part of the dissertation takes a critical look at the way the results of cloud parametrizations are evaluated today. A number of studies using a variety of data sources and modelling approaches are described and the need for a coordinated use of the various existing validation techniques is highlighted. A strategy to achieve such coordination is proposed. This work provides contributions to virtually all facets of the development of cloud parame- trizations. It combines theoretical aspects with the use of a variety of modelling approaches and data sources for the assessment of the performance of the parametrization. All model improvements described here are now part of the operational version of the ECMWF forecast model.