Accurate detection of green-up dates for Leymus chinensis is important to simulate and predict vegetation phenology shifts under the influence of climate change in Inner Mongolia. To identify green-up date of Leymus chinensis steppe effectively, double-Gaussian, Savitzky-Golay, and polynomial functions were used to reconstruct normalized difference vegetation index (NDVI) time series. After evaluations by visual inspection and root mean square error, the most feasible reconstruction method was identified. The maximum slope threshold, dynamic threshold and moving average methods were then used to derive green-up dates from the best reconstructed NDVI time series. Performance of these three methods for green-up date identification was tested against green-up data from 33 ground observation samples and corresponding Moderate-resolution Imaging Spectroradiometer NDVI time-series data at 250-meter resolution. NDVI time series reconstructed with Savitzky-Golay method indicated that green-up dates identified with moving average method agreed well with actually observed ground phenology data.