The modeling and understanding of the deep-water formation variability, especially in the North Atlantic sector, is of crucial importance for the common global ocean variability, in particular on interannual to decadal time-scales. The local restriction of the deep water formation areas makes it necessary to follow new model approaches that are able to resolve these areas with a sufficient high resolution without ignoring the global context. This study aims to validate the ability of the Finite-Element Sea-Ice Ocean Model (FESOM) to reproduce a reliable deep water formation in North Atlantic ocean and to analyse its variability on interannual to decadal time-scales. The FESOM approach works on unstructured triangular surface meshes, which allows us to faithfully resolve coastlines and local areas of interest. The first part of the thesis presents the characteristics of a global FESOM setup designed to study the variability in the deep-water formation areas over five decades for the period 1958-2004. The setup features a regionally increased resolution in the deep water formation areas in the Labrador Sea, Greenland Sea, Weddell Sea and Ross Sea as well as in equatorial and coastal areas. Further, this part of the thesis deals with the applied spinup procedure and the general validation of the FESOM model setup with respect to the performance of the sea-ice and ocean model component. Based on the analysis of the Atlantic Meridional Overturning Circulation (AMOC) we demonstrate that the upper ocean is converged within the applied spinup procedure. The sea ice model reproduces realistic sea-ice distributions and variabilities in the sea ice extent on both hemispheres as well as sea ice transport that compares well with observational data. The general ocean circulation model is validated based on a comparison of the model results with Ocean Weather Ship data in the North Atlantic. We can prove that the vertical structure is well captured in areas with improved resolution. Further, we are able to simulate the decadal ocean variability in the Nordic Sea Overflows as well as several salinity anomaly events and corresponding fingerprint in the vertical hydrography. The second part of the thesis focuses on the validation of the model capability to reproduce a realistic deep-water formation in the Labrador Sea. Therefor, we examine two classes of Labrador Sea water (LSW) which are analysed and compared to observed LSW layer thicknesses derived from profile data for the time interval 1988-2007. We show, that the model setup reproduces in the temporal evolution of the potential density, temperature and salinity two different phase since the late 1980s. These two phases are well known in observational data and are characterized by a significantly different LSW formation. Whereas the first phase features a dominant increase in the layer thickness of the deep Labrador Sea water (dLSW), is the second phase characterized by a degeneration of dLSW. To highlight the processes that are responsible for the variability in dLSW layer thickness we apply a Composite Map Analysis (CMA) between an index of dLSW and sea level pressure, as well as the thermal and haline contributions to the surface density flux. The composite maps reveal that a North Atlantic Oscillation like pattern is one of the main triggers for the variability of LSW formation in the model. Our model results indicate that a massive dLSW formation can act as a low-pass filter to the atmospheric forcing, so that only persistent NAO events correlate with the dLSW index. Additionally our results show that the central Labrador Sea in the model is dominated by the thermal contributions of the surface density flux, while the haline contributions are shielded from the central Labrador Sea by the branch of the Labrador Sea Boundary Current system. In our model, this shielding allows only a minor haline interaction with the central Labrador Sea by lateral mixing. Another aim of the thesis is to examine the general model variability on interannual to decadal time scales. Therefore we study the variability in a normal and random forced FESOM run. By definition of a North Atlantic Deep Water (NADW) index for the normal and random forced FESOM run we could identify an interannual and quasi decadal variability of 7.1 yr and 14.2 yr, respectively. It is found that the normal forced run is dominated by the quasi decadal variability and the random forced run by the interannual variability. The quasi decadal variability could be attributed to the atmospheric forcing, while the interannual variability could be linked to internal modes of the ocean. We defined in analogy to the baroclinic mass transport index (BMT) a DGyre from the horizontal barotropic streamfunction. The comparison of the observed BMT index and the modeled DGyre index reveals that the model is able to reproduce the variability of the index comparing to the observed one, although the model tends to overestimate the magnitude of the index. To further isolate the horizontal but also the vertical variability in the model we apply a principal oscillation pattern (POP) analysis in a three dimensional context. We discovered two exceptional strong interannual modes whose variability could be attributed to a propagating Rossby wave structure.