Atmospheric density models are currently the limiting factor in the accuracy of the dynamic orbit determination and prediction of satellites in a low Earth orbit. Any improvement in these models would greatly aid in applications such as re-entry prediction, ground-track maintenance of Earth observation satellites and forecast of possible collisions with space debris objects. Because of their use in scientific studies, improving these models will also bene?t our understanding of the physical processes that occur in the Earth’s upper atmosphere. Accelerometer instruments onboard of low Earth orbiting satellites are near perfect instruments for studying atmospheric density. They provide accurate observations of the non-gravitational accelerations acting on a satellite and for low flying satellites atmospheric drag is the dominant non-gravitational force. Unfortunately, the number of satellites equipped with an accelerometer is limited. Therefore, in this thesis a strategy is developed and implemented to optimally derive non-gravitational accelerations from precise GPS satellite tracking observations of low flying satellites. This estimation of non-gravitational accelerations using a precise GPS-based orbit determination scheme is referred to as GPS-based accelerometry. The developed GPS-based accelerometry processing strategy is applied to GPS data from the CHAMP, GRACE and GOCE satellites. These satellites carry electrostatic accelerometer instruments, which allows a validation of the GPS-based accelerometry performance. This validation shows that best results are generally obtained for lower flying satellites, due to the increased effect of drag at lower altitudes. In general, best performance is obtained in flight direction, due to the strong effect of accelerations in this direction on orbital dynamics. A comparison with state-of-the-art non-gravitational models shows that for all selected satellites, GPS-based accelerometry outperforms the non-gravitational models in the flight direction. These results indicate that with GPS-based accelerometry, satellites equipped with accurate GPS tracking instrumentation can contribute significantly to atmospheric density modeling. With the growing number of satellites equipped with accurate GPS receivers, this strategy could be applied to a large range of satellites, which offers great potential for the improvement of atmospheric density models.