For the first time, scalings for density peaking are obtained from a database consisting of observations from two devices, ASDEX Upgrade and JET. It is shown that by combining observations from different devices, while some correlations are indeed reduced, also additional uncertainties are introduced. The way which has been adopted in order to overcome the limitations encountered is discussed. Multiple regression analyses are performed which show that in the combined database of ASDEX Upgrade and JET observations, collisionality is the most relevant parameter in the regressions. The particle source provided by neutral beam injection provides a contribution to the peaking, which, despite being non-negligible, remains smaller than 30% and therefore not large enough to explain the whole observed variation of the density peaking. The device size is found to play a small role in scalings which include collisionality, while becomes relevant in scalings which exclude collisionality and include the ratio of the density to the density limit. This indicates that density peaking is more likely to be a function of collisionality rather than of the fraction of the density limit. Scalings for density peaking are proposed and ITER projections are discussed. It is found that all the scalings which include collisionality in the regression variables predict a peaked density pro le for the ITER standard scenario parameters.