ABSTRACTA new source of digital elevation data, the advanced spaceborne thermal emission and reflection radiometer (ASTER) global digital elevation model (GDEM), has been freely available since 2009. It provides enormously greater coverage of the Arctic than previous satellite derived ‘global’ digital elevation models, extending to a latitude of 83 °N in contrast to 60 °N. The GDEM is described as a preliminary, research grade product. This paper investigates its accuracy in a number of specifically Arctic landscapes, including ice and snow, boreal forest, tundra and unvegetated terrain, using test sites in Svalbard, Iceland, Norway and Russia. Semivariogram analysis is used to characterise the magnitude and spatial correlation of errors in the GDEM products from the test sites. The analysis suggests that the horizontal resolution of the GDEM data is around 130 m, somewhat coarser than the sampling interval of 1 second of latitude and longitude. The vertical accuracy is variable, and the factors influencing it have not been systematically explored. However, it appears that the likely accuracy can be estimated from ‘stacking number’ data supplied with the elevation data. The stacking number is the number of independent digital elevation models averaged to generate the supplied product. Provided that this number is greater than around 6 the data have an rms accuracy of typically 5–10 m.