Description: In this thesis a dynamical forecast approach is considered to evaluate the potential seasonal predictability in the European-Atlantic region with emphasis on the mean winter climate. Two state-of-the-art seasonal forecast systems are used, namely the Seasonal Forecast System 2 from the European Centre for Medium Range Weather Forecast (ECMWF) and a multi-model system developed within the joint European project DEMETER (Development of a European Multi-Model Ensemble Prediction System for Seasonal to Interannual Prediction). The predictions are verified with the ERA-40 re-analysis data. Seasonal forecasts are probabilistic in nature and hence require verification techniques based on probabilistic skill measures. Here a multi-category skill score, namely the ranked probability skill score (RPSS) is applied. The RPSS is sensitive to the shape and the shift of the predicted probability density distribution. However, the RPSS shows a negative bias for ensemble systems with small ensemble sizes. It is shown that the negative bias can be attributed to a discretization and squaring error in the quadratic norm of the RPSS. In the following two strategies are explored to tackle this flaw. First, it is shown that the RPSSl=i based on the absolute rather than the squared norm is unbiased. Nevertheless, it is not strictly proper in a statistical sense. Second, an unbiased and strictly proper skill score can be defined based on the quadratic norm, along with the reference forecast reduced to sub-samples of the same size as the forecast ensemble size. This is denoted as the de-biased ranked probability skill score (RPSSd). Based on a hypothetical set up comparable to the ECMWF hindcast system (40 members, 15 hindcast years) the RPSSd is used to show, that statistically significant skill scores can only be found for climate anomalies with a signal-to-noise ratio larger than -0.3. Furthermore, the seasonal predictability is evaluated using a forecast approach (FA) based on 2m mean temperature predictions on grid-point scale for the years 1987-2001. The ECMWF Seasonal Forecast System 2 provides a marked improvement in skill relative to climatological forecasts over the North-Atlantic Ocean with maximum values of up to 30 %. Over Europe no significantly positive skill scores are found. The DEMETER multi-model has higher forecast skills than individual models. Moreover, the potential predictability is investigated applying a perfect model approach (PMA). Such approach assumes that the climate system is fully represented by the model physics. The potential winter predictability over the European continent amounts to approximately -10%. The 3r part of the thesis examines the potential seasonal predictability is examined via the leading mode of the European winter climate variability, namely the North Atlantic Oscillation (NAO). The PMA shows that the mean winter NAO and the NAO temperature related impact is potentially predictable for lead time 1 month, but with a gain in skill of only -8 % compared to climatology. Using the FA, the results are quite different. For the period 1959-2001, the NAO skill score is not statistically significant, while the skill score is surprisingly large (16 % to 27 % relative to the observed climatology) for the period 1987-2001. For this period a weak relation between the strength of the NAO amplitude and the skill score of the NAO is found. This contrasts with ENSO variability where the amplitude dependent forecast skill is strong. Finally, the seasonal forecasts are examined from the end user's perspective. A so-called "Klimagram" is introduced to assess seasonal climate forecasts for particular cities or regions. A first analysis reveals that the forecast skills can be improved in a relative sense, looking at spatial and temporal averaged quantities. Overall, this study suggests a positive potential seasonal predictability in the European-Atlantic domain in winter. However, the potential benefit is rather small and constitutes a fraction only, compared to currently possible results in the tropics.