OHP lidar data and National Centers for Environmental Prediction (NCEP) stratospheric temperature analyses provide long and continuous databases for the middle and upper stratosphere that are highly valuable for long-term studies. However, each data set has limitations. Comparisons between lidar data from 1979 to 1993 and NCEP data interpolated from the global analyses to the lidar location reveal significant mean temperature differences. Insight into the origin of the differences offers an opportunity to improve the overall quality of temperature monitoring in the stratosphere. Some of the differences can be explained by instrumental effects in the lidar system. In the stratosphere most of the limitations in lidar temperatures appear below 35--40 km, due to events of lidar misalignment (as large as 10 K) or to the effects on lidar data of volcanic aerosols (as large as 15 K). Changing biases between lidar and NCEP temperatures above 5 hPa coincide with replacement of satellites used in the NCEP analyses. However, some bias differences in upper stratospheric temperatures remain even after NCEP adjustments are made, based on rocketsonde comparisons. While these biases have been already suspected, they had never been explained. Here we suggest that the remaining bias (2--4 K) is caused by tidal influences, heretofore not accounted for by the NCEP adjustment procedure. Lidar profiles have been filtered in their lower part for misalignment and aerosol contamination. Long-term changes have been compared, and a factor of 2 in trend differences have been reported. No significant trends (at 95% confidence) have been detected except with lidar around the stratopause and with NCEP analyses at 5 and 10 hPa. According to instrumental limitations of both data sets the temperature trend may vary from 1 to 3 K with altitude (10--0.4 hPa). Because only satellite data can provide global trend estimates and because lidar data have been chosen for ground-based stratospheric monitoring programs, we suggest some plans to overcome these difficulties for past and future measurements. This should allow a more confident use for future trend estimates from both data sets. ¿ 2001 American Geophysical Union