Abstract: Based on GIMMS 3g (the third generation Global Inventory Modeling and Mapping Studies Normalized Difference Vegetation Index) data, three preprocessing methods including monthly aggregation, standard anomaly computation, and trend-preserving prewhitening were used to develop six data series. Seasonal trend analysis was applied to extract three seasonal representative factors, i.e. amplitude 0, amplitude 1, and phase 1 to detect the characteristic of seasonal trend. The interannual and seasonal trend analysis was conducted using CMK (Contextual Mann-Kendall) and MK (Mann-Kendall) trend test methods. Land use and cover data was used as an accessory to identify spatial distribution pattern of areas with significant changes. The difference between preprocessing method and trend test method was also discussed. The result shows that:(1) The proportional area of vegetation deterioration is higher than the proportional area of vegetation improvement; the former is mainly located at areas of unused lands and grasslands, and the latter is found over grasslands, unused lands and farmlands. (2) The amplitudes of annual variability show a significant increasing trend mainly over grasslands, unused lands and farmlands in the southern margin of the Tarim basin. (3) Different preprocessing methods have obvious impacts on the result of trend analysis. According to the ability of these methods to extract significant trend information, they are in the sequence of standard anomaly > trend-preserving prewhiting > original data > monthly aggregate. (4) 87.88% of farmlands demonstrates a significant increase trend in annual variation amplitude and 53.31% of farmlands shows a significant trend of delayed onset of growing season.