Firn cores from different locations in North Greenland are investigated to analyse the influence of the accumulation rate and the impurities on the densification. The drilling regions have similar mean annual temperatures, but significantly different rates of ac-cumulation. In this study the density profiles of three firn cores are measured with the new established method of X-ray radioscopy. These and six other density profiles from previous studies are used to find some general pattern of densification in North Green-land. The influence of impurities on densification are analysed by considering the calci-um and dust concentrations for one firn core measured in the framework of an ongoing PhD-work and by measuring also the sodium, nitrate and ammonium concentrations for particular sections of the three firn cores with the means of continuous flow analysis. The classic densification model by Herron and Langway [1980] and the model by Freitag et al [2013], which is extended with the effect of impurities, are used to simulate the density and the density variability. It is shown that in North Greenland the densification of firn is different for high and low accumulation rates. The accumulation rate affects the depth profile of the density variability leading to profiles with a minimum and a second maximum at high and to uniform plateau-like profiles at low accumulation rates. This implies that for regions with high accumulation rate a strong phase relationship between the density and the dust concentration at the snow surface exists, which is inversed. Consequently, the density at the snow surface has a seasonal signal. In contrast, regions with lower accumulation rate have no strong phase relationship, so that the density of the snow surface has no seasonal behavior. With increasing depth the density follows the seasonal depth profile of the impurities. Both, the model by Herron and Langway [1980] and the model by Freitag et al. [2013] simulate the mean density profile, whereas the density variability was only reproduced by the extended model of Freitag et al. [2013].