The database obtained by Rayleigh lidars over the south of France is now used for monitoring the middle-atmosphere structure and to validate satellite data. For these reasons it is crucial to ensure the quality of the data. The purpose of this paper is to review all possible sources of errors that could induce random or systematic bias in the temperature measurements. The characteristics of the lidars, the procedures used, as well as the data reduction software are then reviewed. Comparisons made between the lidar and other available techniques and between lidars of different characteristics lead to the conclusion that an accuracy of 1 K can be attained between 30 and about 70 km depending on the lidar power. The method itself is not affected by drift with time and provides absolute temperature data without any need of calibration and therefore is one of the best instruments for long-term monitoring.