In many GNSS software packages a simplified observation weighting model is used which is merely based on the satellite elevation angle and valid under the assumption of azimuthal symmetry. This elevation-dependent weighting model is only suitable for undisturbed GNSS signals based on the existing strong correlation between signal quality and satellite elevation angle. However, for high-precision geodetic applications this geometry-related weighting model becomes obsolete if observations are strongly affected by multipath effects, signal diffraction as well as receiver characteristic under non-ideal observation conditions. An improved observation weighting model based on signal-to noise power ratio measurements has been developed and experimentally implemented in the Bernese GPS software 5.0. Tests indicate that when this weighting model is used for low elevation data additional 10% ambiguities can be resolved and the accuracy of the estimated site-specific neutrosphere parameters can be improved by nearly 25% compared with the standard elevation-dependent weighting model