Abstract: Using 445 temporal AMSR-E brightness temperature and snow depth data measured by 20 meteorological stations during the three snow seasons from December to March in 2002 to 2004 in Northern Xinjiang, we systematically analyzed the approaches of screening samples and impact factors for establishing a snow depth model. Through regression analysis of horizontal and vertical polarization brightness temperature differences in the 18 and 36 GHz bands, and of snow depth values, a snow depth model was established based on the AMSR-E brightness temperature data in northern Xinjiang. Accuracy of the model was evaluated: 1) AMSR-E brightness temperature differences were seriously affected by temperature, snowmelt, rain, wet snow and deep frost layers, of which the most important factor was deep frost layers. 2) There was a significant correlation between snow depth (y) over 2.5 cm and the vertical polarization brightness temperature difference at 18 and 36 GHz. The equation was SD=0.49(Tb_18V－Tb_36V)+8.72, and the correlation coefficient was up to 0.65. 3) The average error was -7.1 cm and average absolute error was 7.1 cm and RMSE was 7.7 cm when snow depth was 3-10 cm. When snow depth was between 11 and 30 cm, the average error was 1.8 cm, average absolute error was 4.9 cm, and RMSE was 9.1 cm. When the snow depth was over 30 cm, average error was 8.9 cm, average absolute error was 9.4 cm, and RMSE was 18.1 cm. 4) The snow depth model is more accurate than the Chang algorithm in Northern Xinjiang. It showed a good agreement between the simulated and measured snow depth values when the surface was covered by middle depth snow. However, the model’s accuracy was lower when the surface was covered by fallow or deep snow. Basically, the model can reflect the trend of snow depth variation in Northern Xinjiang, but it has a low accuracy for fallow or deep snow, and needs to be improved in the future.