The objective of this study is to investigate the impact on temperature forecast of using ozone analyses for the computation of heating rates in a three-dimensional variational data assimilation (3D-Var) system with a coupled model. The system is based on a tropospheric-stratospheric forecast model that includes a comprehensive stratospheric chemistry module for online resolution of the dynamical, radiative, and photochemical interactions. The system assimilates conventional observations as well as temperature and ozone measurements from the Michelson Interferometer for Passive Atmospheric Sounding (MIPAS) instrument. Several data assimilation cycles have been performed over the period August-October 2003 to produce a set of analyses that have been used for launching an ensemble of 10-day forecasts. Temperature and ozone forecasts have been compared with MIPAS and radiosonde observations in different regions. Results show that, in the absence of ozone assimilation, the impact of using a prognostic ozone distribution for the computation of heating rates as opposed to monthly mean climatologies is generally neutral. With the addition of ozone assimilation, the improvement against a noninteractive assimilation system is systematic and occurs over a wide range of time scales throughout the lower stratosphere. The improvement on 6-h temperature forecasts is mainly seen in the Southern Hemisphere, where ozone analyses are in good agreement with observations. For 10-day forecasts, the impact of using ozone analyses is more important in the Northern Hemisphere, where it improves the temperature predictability by more than 1 day at 50 hPa. Comparisons with analyses also show a systematic reduction of the temperature root-mean-square errors and biases throughout the assimilation period. The overall results demonstrate that a comprehensive coupled 3D-Var system that incorporates the radiative feedback from ozone analyses can be used for improving temperature predictability throughout the stratosphere. Comprehensive approaches can be used as a benchmark for the development of linearized methods for improving temperature and ozone forecasting in the region. © 2009 American Meteorological Society.