Temporal changes in the normalized difference vegetation index (NDVI) have been widely used in vegetation mapping due to the usefulness of NDVI datasets in distinguishing characteristic seasonal differences in the phenology of greenness of vegetation cover. The Timeseries Moderate Resolution Imaging Spectroradiometer (MODIS) NDVI datasets hold considerable promise for large-area land cover classification given their global coverage, intermediate spatial resolution, high temporal resolution (16=day composite period), and cost-free status. This study focused on generating effective classification features from multi-temporal MODIS NDVI datasets to improve classification accuracy in the Heihe River Basin. Two types of features were derived from reconstructed multi-temporal MODIS NDVI datasets. The first are the basic parameters including the annual maximum NDVI, the mean NDVI during the growing season, the inter-annual variability of NDVI and the annual mean NDVI. The second are the amplitude and phase information of the first and second harmonic components derived from the shape of the time-series NDVI profile. Additionally, DEM with 1km resolution has also been used to simplify the current scheme. According to the validated results with 469 ground truth survey samples, the overall land cover classification accuracy using the decision tree was 78% and a Kappa coefficient is 0.74. The results support using decision tree classification based on 1km MODIS NDVI temporal and derived parameters to provide an up-to-date land cover mapping. However, the current decision tree does not work well in the downstream of  the Heihe River Basin since the NDVI of non-vegetation types can not represent the temporal feature of these types. Thus, new effort is necessary in the future in order to improve the overall performance on this issue.