Abstract: As a data source needed in many research fields，the importance of the ground elevation data is obvious. As far as the free worldwide ground elevation data are concerned，there would be great shortage in the analysis of the data errors. The data errors are much more obvious especially in the alpine regions. There are lots of articles about the accuracy assessment and error analysis for the ground elevation data at home and abroad，but only very few articles analyze the errors of ground elevation data in the alpine regions. There is still no detailed scientific assessment about the accuracy level and error situations of the ground elevation data in the alpine regions and mountainous areas. With the DEM by a scale of 1∶50 000 in the Tianshan Mountains regions of Xinjiang，China as the benchmark data，and taking the geological factors such as gradient，exposure，hillslope and relief amplitude into consideration，the article uses three numerical precision models of Mathematical and Statistical Analysis including [RMSE] （root mean square error），ME （mean error） and [SD] （standard deviation） to fully analyze and compare the error situations of the three digital elevation models （SRTM，ASTER GDEM，ICESat GLAS14） which are commonly used at home and abroad in the alpine regions in detail.  The results indicated as follows：firstly，the three ground elevation data were affected by the gradient to the different degrees. The gradient has the most effect on the SRTM，and the least effect on ASTER GDEM.  In the flatter terrain area with a gradient less than 20°，the accuracies of SRTM and ICESat GLAS14 are higher than that of ASTER GDEM，and the accuracy of ICESat GLAS14 is higher than that of the others. What’s more，when in the flat terrain area with the gradient between 0°~10°，the average error is only 1.231 m. And when in the steep mountain areas with the gradient between 20°~45°，the accuracies of SRTM and ICESat GLAS14 are lowered to different degrees，and the lowering speed of SRTM is the fastest，the [RMSE] of which reaches up to 40 m，which is lower than that of ASTER GDEM. And in the steep mountain area with the gradient greater than 45°，the accuracies of SRTM and ICESat GLAS14 are lowered continuously，even the accuracy of ASTER GDEM is higher than that of ICESat GLAS14. Secondly，the similar accuracy levels of ICESat GLAS14 in different exposures indicate that the ICESat GLAS14 is less affected by the exposure. The accuracy levels of SRTM and ASTER GDEM are constrained to different degrees because of the differences of data access and data coverage. Thirdly，with the increasing of the relief amplitude，the RMSE of the three ground elevation data presents a growing trend. Among them，the largest increase of the RMSE of SRTM approaches to 35 m and the next ones are the ICESat GLAS14 and ASTER GDEM. When in a range of the topographical height differences between 35 m and 50 m，the RMSE levels of the three ground elevation data are similar. And when beyond the range，the RMSE level of SRTM is higher than that of ASTER GDEM. The effects of the relief amplitude and gradient on the two elevation data differences are similar，because the regions with great topographic relief usually correspond to the regions with steeper gradient.  The innovative points of the article are as follows: data statistics are carried out with numerous experimental data points （153 282 points），and the error analysis of the three ground elevation data are based on local absolute accuracy and the overall relative accuracy. The aim of the article is to make a conclusion which is suitable for the analysis of the errors of ground elevation data in the alpine regions and could provide a scientific reference to the application of the ground elevation data in alpine and mountainous areas.

Key words: DEM, alpine areas, error analysis