This paper presents a chemical assimilation system based on the variational method. It has been applied to an off-line stratospheric three-dimensional chemical transport model. This technique determines the best model initial conditions that minimize the model errors when compared to a set of observations for a predefined time window. The system is applied to the Cryogenic Infrared Spectrometers and Telescopes for the Atmosphere (CRISTA) measurements for the November 5--11, 1994, mission. Observed species are O3, CH4, HNO3, ClONO2, N2O5, N2O, and CFC-11. For an assimilation period of 12 hours, CRISTA observations are well estimated by the assimilation system. These results vary according to the species considered and depend on the CRISTA's observational error. In particular, assimilated ozone differ from CRISTA by less than 5%. Analyses have been compared with independent observations in order to validate the assimilation system. It was found that observations by the Atmospheric Trace Molecule Spectroscopy (ATMOS) experiment and the Halogen Occultation Experiment (HALOE) agree well with the analyses. These comparisons suggested systematic differences between CRISTA and HALOE or ATMOS. Unconstrained constituents like NOx and HCl are influenced through chemical coupling and show small discrepancies with ATMOS and HALOE data. ¿ 2001 American Geophysical Union