16 K in 1988-1989, to +0.26 K in 1994-1996 and −0.46 K after 1998, the latter being fully consistent with recent evaluations of ECMWF operational analysis from radio occultation and in situ longduration balloon observations. The amplitude of the bias and its evolution are very similar to the results of a previous evaluation from radiosondes in 1991-2003, suggesting that the origin of the drift of ERA-40 might be mainly due to errors in the series of satellite measurements of MSU, replaced by AMSU in 1998, assimilated in the model.The ERA-40 zonal wind speed in the lower stratosphere appears slightly overestimated by 0.7-1.0 m/s on average in both the tropics and equatorial region, that is by 5-10% compared to the average 10-20 m/s wind speed. This bias, fully consistent with a recent evaluation of ECMWF operational analysis in 2004, is found constant during the whole 1988-2001 period, suggesting a shortfall in the variabililty of ERA-40 horizontal winds in the lower stratosphere in the tropics and the equatorial region. Finally calculated trajectories using ERA-40, frequently used for analysing field observaCorrespondence to: T. Christensen (tic@dmi.dk) tions, are found in error compared to that of the balloons by ±500 km after 5 days and ±1000 km after 10 days.