We have used a three-dimensional stratospheric model to investigate Arctic ozone depletion in the three winters from 1995/1996 to 1997/1998, winters with quite different dynamical conditions. We have diagnosed the chemical depletion from the model and compared the model ozone with ground-based and sonde measurements. In winter 1997/1998 the lower stratosphere was generally much warmer than the previous two cold winters. However, the occurrence of low temperatures in December 1997 did cause chlorine activation and ozone loss in the model. In fact, despite the very different meteorological conditions for these three winters, by mid-February the model-calculated mean vortex loss was similar in each year at around 20% at 480 K. Larger differences in the calculated loss occurred when low temperatures persisted into March; this seems a prerequisite for very large loss. By late March the model-calculated mean vortex column loss was 53 Dobson units (DU) in 1998, 71 DU in 1996, and 80 DU in 1997. A first effort at a detailed day-to-day validation of a chemical transport model by comparison with ozone measurements is presented. The model reproduces the mean column observations to within about ¿10%, although the bias varies in magnitude and changes sign in different years. The agreement between the model and ozonesonde profiles in the lower stratosphere varies from year to year and during the course of a particular winter/spring. In winter 1997/1998, the model agrees well with sonde profiles at the 480 K and 555 K levels with a positive mean difference (ozonesonde values minus model values) with amplitude of less than 6%. In contrast, in winter 1996/1997, the mean difference is negative and the amplitude is less than 15%. In winter 1995/1996 the mean difference is negative and the amplitude reaches 43%. The differences in 1995/1996 and 1996/1997 result despite good agreement between model and observations in midwinter, and they develop following the large chemical losses which occurred in those winters. We suggest that a major contribution to the model/observation differences for those two winters is a model underestimation of the chemical ozone loss. ¿ 2000 American Geophysical Union