Global Positioning System (GPS) measurements have been demonstrated to provide precipitable water vapor (PWV) estimates with a level of accuracy that is comparable to that of radiosondes and microwave radiometers. GPS measurements therefore have the potential to become a significant source of data for operational weather forecasting, provided that PWV (or the intermediate zenith total delay (ZTD)) can be made available in near real-time with a minimum accuracy degradation. Despite the recent decrease in the latency and increase in accuracy provided by the International GPS Service (IGS) ultrarapid predicted GPS orbit products, we show that the accuracy of these orbits continues to be a limiting factor for the accuracy of near real-time GPS-derived atmospheric estimates. In this work, a coefficient matrix is derived from the normal equations of the least squares adjustment model for the GPS observables that maps the orbital parameter errors into ZTD errors. This is used to analyze the sensitivity of GPS derived tropospheric errors to an extensive set of parameters, including their time dependence, in a computationally efficient manner. We show that ZTD errors are dominated by biases in the orbital semimajor axis, with minor contributions from the inclination and argument of perigee, and that this error increases significantly after the fourth to fifth hour of the prediction window. We implemented a GPS data processing strategy based on an iterative estimation of the three most critical orbital parameters (semimajor axis, inclination and argument of perigee) together with the ZTD parameters. We tested this strategy in a 3500 ¿ 3500 km network of 15 GPS sites in western Europe providing hourly data files. We show that the standard deviation improvement compared to a strategy based only on the orbit quality index provided with the predicted orbit products is on the order of 20%. The analysis of one month of data in near-real-time shows a bias lower than 1 mm ZTD and a standard deviation lower than 6 mm ZTD compared to using the most precise IGS final orbits. We also show that this strategy is robust and capable of dealing with very large orbit errors appropriately. We demonstrate that the same quality is achievable with a 1500 ¿ 1500 km network which has positive implications for decentralized processing strategies. The near real-time processing methodology described here meets the current timeliness requirements of operational meteorology (30 mn to 2 hours, depending on the application), while ensuring a level of accuracy similar to that provided in postprocessed mode with precise final IGS orbits (1 mm ZTD bias, 6 mm ZTD RMS). The method we propose can also be considered as an on-the-fly orbit quality control for near real-time GPS applications.