Abstract: Loess landform in Xinjiang and interpretation of its remote-sensed images are somewhat different from those in the Loess Plateau. Due to the special and complex physical geographical environment of Xinjiang， there are some difficulties sometimes in interpreting loess landform in remote-sensed images. According to the author’s practice and knowledge from experience in the project of 1∶1 000 000 digital geomorphic mapping of China （part Xinjiang），based on geographical related analysis method， field on-site inspection and validation method， vector editing tools and spatial analysis tools of GIS software， and using a series of Xinjiang’s data including digital geomorphic map（1∶1 000 000），DEM（1∶250 000），TM/ETM and Google Earth remote-sensed images，vegetation map（1∶1 000 000），soil map（1：1 000 000） and land use/land cover map（1∶100 000）（2000），this paper puts forward seven bases for the remote sensing interpretation of loess landform in Xinjiang，which deal with such aspects as loess landform’s location，tone and color，size，shape，shadow，drainage and texture，vegetation and soil，related object in the remote-sensed images，and related material and field experience. Furthermore，some identification characteristics between loess landform and sandy desert landform，arid landform，fluvial landform，periglacial landform，yardang landform and some other special circumstances in Xinjiang，are summed up. The paper points out that： loess landform bodies of Xinjiang usually connect with the mountains and lie on the outside edge of the mountains，and commonly are loess hills or outer mountains covered with loess，for loess often overlaps on other terrain as a cover layer； in remote sensing images，except for the valleys，the overall color of loess landform is single and uniform，its tone and shade transition is generally gentle and gradational，and the tones of loess landform in areas of desert，desert steppe and typical steppe，are totally light and pale（but generally not dazzling bright white），often with oily gloss and obvious differences from surrounding landforms； generally distributing as sheets，loess landform bodies of Xinjiang are usually easy to be made out due to their big image areas； the ridges of loess landform bodies are normally even and without sharp feeling，the transition of slopes is smooth and gradational，the shadows of shady slopes and valleys are lightish，and the valleys are wide and shallow or narrow but not deep； compared with loess landform in the Loess Plateau，the development of valleys in loess landform areas of Xinjiang is significantly weaker，the depths of valleys are shallower and the valleys mainly are shallow grooves and ditches，the density of gullies and ravines are apparently less，and the shades of valleys in images are distinctly lighter； with less or absence of main valleys，or with unclear levels between main and second valleys，the arrangement of same level valleys in loess landform areas of Xinjiang is orderly relatively，the sunny sides of valleys often form long and pan-white stripes，the valley lines curve gently and soft，the combination patterns of valleys are commonly like plaits，approximate parallels，feathers or radii，or not obvious due to the influence of some factors such as control of underlying terrain. Besides，vegetation types，soil types，related objects，and other related materials as well as interpreter’s field experience also can be sound bases for interpretation.Moreover，a viewpoint is given for the first time that in many loess landform areas of northern Xinjiang，the light to deep color like red ochre，which is presented in the images synthesized by 7，4，2 bands of TM or ETM，can be caused by the reddish vegetation or reddish litter，and thus it needs more carefulness to differentiate loess landform from exposed red rocks and those non-loess landforms which also have reddish vegetation or reddish litter on the ground.The above results may provide some reference for the future geomorphic mapping of Xinjiang and for other related scientific researches and teaching work.

Key words: Xinjiang, loess landform, remote sensing interpretation, geomorphic mapping