To evaluate the quality and temporal consistency of the large-scale Brewer-Dobson circulation in meteorological data from the European Centre for Medium-Range Weather Forecasts, the stratosphere-troposphere exchange of ozone is calculated using a model with linearized stratospheric ozone chemistry. Simulations are performed with the 45-year reanalysis ERA-40 and with operational data (OD) for the period November 1999 to March 2005. In both hemispheres the ozone exchange fluxes are generally higher with ERA-40 than OD. In the Northern Hemisphere this discrepancy appears to be exacerbated by the introduction of satellite observations into the ERA-40 assimilation system in 1973. A comparison to a simulation with data from a run of the ERA-40 system in which no radiance observations were assimilated for the period January--March 1973 serves to demonstrate this point. The ozone exchange flux was also analyzed as a function of the forecast range of the assimilated winds used to drive the model. For optimal comparison between the two assimilation systems, this is done for the year July 2001 to June 2002, when there is maximum overlap between the respective model cycles. The sensitivity to forecast time and update frequency of the forecast series is found to be much stronger for ERA-40 than OD. For the later years of the reanalysis it is concluded that the circulation bias introduced by the assimilation of (satellite) observations is strongly reduced during forecast mode. From a chemistry transport modeling perspective the results imply that the stratospheric and upper tropospheric transport of tracers like ozone can be improved by using the forecasts of the reanalysis but is best represented with the operational data.