Plant phenology plays a significant role in regulating carbon sequestration period of terrestrial ecosystems. Remote sensing of land surface phenology (LSP), i.e., the start and the end of the growing season (SOS and EOS, respectively) in evergreen needleleaf forests is particularly challenging due to their limited seasonal vari- ability in canopy greenness. Using 107 site-years of CO2 flux data at 14 evergreen needleleaf forest sites in North America, we developed a new model to estimate SOS and EOS based entirely on the Moderate Resolution Imaging Spectroradiometer (MODIS) data. We found that the commonly used vegetation indices (VI), including the normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI), were not able to detect SOS and EOS in these forests. The MODIS land surface temperature (LST) showed better performance in the es- timation of SOS than did a single VI. Interestingly, the variability of LST (i.e., the coefficient of variation, CV_LST) was more useful than LST itself in detecting changes in forest LSP. Therefore, a new model using the product of VI and CV_LST was developed and it significantly improved the representation of LSP with mean errors of 11.7 and 5.6 days for SOS and EOS, respectively. Further validation at five sites in the Long Term Ecological Research network (LTER) using camera data also indicated the applicability of the new approach. These results suggest that temperature variability plays a previously overlooked role in phenological modeling, and a combi- nation of canopy greenness and temperature could be a useful way to enhance the estimation of evergreen needleleaf forest phenology of future ecosystem models.