GNSS observations weighting models have received attention in the past decades. Several reasons can explain this gain in consideration. It mainly corresponds to the emergence of autonomous, global, continuous, precise and integrity-compliant localization systems. Indeed, for safety critical applications, adopting a naive strategy of observations' equal weighting could not guarantee to achieve all goals, and even more for terrestrial applications where the local environment can strongly disturb signals reception. The existing observation models use one or two metrics to estimate the confidence to be given to an observation. The first is the satellites' elevations. This model considers that the satellites with low elevation are more subject to global errors (orbit and atmosphere errors) or local multipaths or NLOS phenomena. The second criterion is the carrier to noise ratio provided by the receiver. However, combining these two criteria by a simple multiplication of the two models taken separately, presents a limit: the result is no more « under control ». In this study, we propose to formulate a new weighting strategy using these two criteria more guided by expertise's expected results.