The　variations　of　the　amount　of　aspect　information　derived　from　a　digital　elevation　model　(DEM)　with　the　DEM＇　resolution　were　studied　using　the　non-linear　regression　method.　In　the　northwest　China＇s　Loess　Plateau,　seven　major　sample　areas　representing　different　landforms　were　selected,　each　of　which　was　divided　into　1.6　subareas.　The　DEMs　with　the　resolutions　of　5m,　15m,　25m,　35m,　45m,　55m,　65m,　75m,　respectively,　were　established　according　to　the　map　scaled　at　1:10000　on　each　sample　area.　The　aspect　information　derived　from　the　DEM　with　the　resolution　of　5m　was　assumed　as　the　true　value,　and　the　information　derived　from　other　DEMs　was　taken　as　the　investigated　subject.　The　functional　relations　between　aspect　information　losses　and　resolutions　were　analyzed.　The　indexes　of　the　aspect　accuracy　(a)　of　different　categories　and　the　information　loss　(q)　based　on　the　single　grid　were　proposed.　In　each　area,　the　function　of　the　indexes　and　different　resolutions　was　obtained　so　that　the　needed　resolution　can　be　calculated　if　the　index　is　known.　The　function　is　helpful　in　selecting　proper　resolutions,　minimizing　the　amount　of　data　and　cutting　down　the　cost　while　ensuring　the　accuracy　of　aspect　information.