In this talk, we discuss diagnostics to measure the efficiency of an ensemble of analyses and ensuing forecasts in representing the error space. We demonstrate the use of these diagnostics by preparing ensembles of analyses and forecasts with a reduced resolution version of the NCEP GFS model. Based on the results of these experiments, we argue that the proposed diagnostics provide a powerful set of tools to diagnose the contribution of model uncertainty to the forecast uncertainty. In particular, we demonstrate that in the medium range and beyond (for forecast times longer than about 48-72-h) in the extratropics, model uncertainty has only modest effects on the dominant error patterns, which are primarily determined by the internal instabilities of the atmospheric flow. Model uncertainty, however, plays an important role in the uncertainty of the magnitude of the different error patterns, which makes predicting the relative importance of the different error patterns by an ensemble challenging. The significance of our findings for statistical post-processing of ensemble forecasts will also be discussed.