AbstractThe Constellation Observing System for Meteorology Ionosphere and Climate (COSMIC) is a six‐satellite Global Positioning System (GPS) radio occultation (RO) mission that started in April 2006. The close proximity of these satellites during some months after launch provided a unique opportunity to evaluate the precision of GPS RO temperature and refractivity profile retrievals in the neutral atmosphere from nearly collocated and simultaneous observations. In order to work with nearly homogeneous sets, data are divided into five groups according to latitude bands during 20 days of July. For all latitude bands and variables, the best precision values (about 0.1%) are found somewhere between 8 and 25 km height. In general, we find that precision degrades significantly with height above 30 km and its performance becomes there worse than 1%. Temperature precision assessment has been generally excluded in previous studies. Refractivity has here, in general, a precision similar to dry temperature but worse than wet temperature in the lower atmosphere and above 30 km. However, it has been shown that the better performance of wet temperature is an artificial effect produced by the use of the same background information in nearly collocated wet retrievals. Performance in refractivity around 1% is found in the Northern Hemisphere at the lowest heights and significantly worse in the southern polar zone above 30 km. There is no strong dependence of the estimated precision in terms of height on day and night, on latitude, on season, or on the homogeneity degree of each group of profiles. This reinforces the usual claim that GPS RO precision is independent of the atmospheric conditions. The roughly 0.1% precision in the 8–25 km height interval should suffice to distinguish between day and night average values, but no significant differences are found through a Student t test for both populations at all heights in each latitude band. It was then shown that the present spatial density of GPS RO does not allow to analyze smaller latitudinal bands, which could lead to smaller dispersions associated with the day and night means, where it would then be potentially possible to detect significant statistical differences among both categories. We studied the uncertainties associated with the background conditions used in the retrievals and found that their contribution is negligible at all latitudes and heights. However, they force an artificial improvement of wet temperature precision as compared to the dry counterpart at the lowest and highest altitudes studied. In addition, we showed that there is no detectable dubious behavior of COSMIC data prior to day 194 of year 2006 as warned by the data providers, but our result applies only to the precision issue and cannot be extended to other features of data quality. Regarding accuracy, we estimated an average bias of 0.1 K for GPS RO temperature between about 10 and 30 km height and somewhat larger at lower altitudes. We expect a roughly −0.5 K bias above 35 km altitude. Regarding refractivity, a −0.2% bias of the measurements was estimated below about 8 km height.