We report results from a multiple linear regression analysis of long-term total ozone observations (1979 to 2000, by TOMS/SBUV), of temperature reanalyses (1958 to 2000, NCEP), and of two chemistry-climate model simulations (1960 to 1999, by ECHAM4.L39(DLR)/CHEM (=E39/C), and MAECHAM4-CHEM). The model runs are transient experiments, where observed sea surface temperatures, increasing source gas concentrations (CO₂, <i>CFC</i>s, CH₄, N₂O, NO_x), 11-year solar cycle, volcanic aerosols and the quasi-biennial oscillation (QBO) are all accounted for. MAECHAM4-CHEM covers the atmosphere from the surface up to 0.01&nbsp;hPa (&asymp;80&nbsp;km). For a proper representation of middle atmosphere (MA) dynamics, it includes a parametrization for momentum deposition by dissipating gravity wave spectra. E39/C, on the other hand, has its top layer centered at 10&nbsp;hPa (&asymp;30&nbsp;km). It is targeted on processes near the tropopause, and has more levels in this region. Despite some problems, both models generally reproduce the observed amplitudes and much of the observed low-latitude patterns of the various modes of interannual variability in total ozone and lower stratospheric temperature. In most aspects MAECHAM4-CHEM performs slightly better than E39/C. MAECHAM4-CHEM overestimates the long-term decline of total ozone, whereas underestimates the decline over Antarctica and at northern mid-latitudes. The true long-term decline in winter and spring above the Arctic may be underestimated by a lack of TOMS/SBUV observations in winter, particularly in the cold 1990s. Main contributions to the observed interannual variations of total ozone and lower stratospheric temperature at 50&nbsp;hPa come from a linear trend (up to -10&nbsp;DU/decade at high northern latitudes, up to -40&nbsp;DU/decade at high southern latitudes, and around -0.7&nbsp;K/decade over much of the globe), from the intensity of the polar vortices (more than 40&nbsp;DU, or 8&nbsp;K peak to peak), the QBO (up to 20&nbsp;DU, or 2&nbsp;K peak to peak), and from tropospheric weather (up to 20&nbsp;DU, or 2&nbsp;K peak to peak). Smaller variations are related to the 11-year solar cycle (generally less than 15&nbsp;DU, or 1&nbsp;K), or to ENSO (up to 10&nbsp;DU, or 1&nbsp;K). These observed variations are replicated well in the simulations. Volcanic eruptions have resulted in sporadic changes (up to -30&nbsp;DU, or +3&nbsp;K). At low latitudes, patterns are zonally symmetric. At higher latitudes, however, strong, zonally non-symmetric signals are found close to the Aleutian Islands or south of Australia. Such asymmetric features appear in the model runs as well, but often at different longitudes than in the observations. The results point to a key role of the zonally asymmetric Aleutian (or Australian) stratospheric anti-cyclones for interannual variations at high-latitudes, and for coupling between polar vortex strength, QBO, 11-year solar cycle and ENSO.