The phenology of vegetation exerts an important control over the terrestrial ecosystem carbon (C) cycle. Remote sensing of key phenological phases in forests (e.g., the spring onset and autumn end of growing season) remains challenging due to noise in time series and the limited seasonal variation of canopy greenness in evergreen forests. Using 94 site-years of C flux data from four deciduous broadleaf forests (DBF) and six evergreen needleleaf forests (ENF) in North America, we examine whether growing season phenology can be remotely sensed from mean vegetation indices (VIs) derived from spring (Apr.–May) and autumn (Sep.–Nov) observations. Five VIs were used based on Moderate Resolution Imaging Spectroradiometer (MODIS) data, including the normalized difference vegetation index (NDVI), the land surface water index (LSWI), the enhanced vegetation index (EVI), the wide dynamic range vegetation index (WDRVI) and the optimized soil-adjusted vegetation index (OSAVI). Our results show that growing season transitions can be inferred from mean seasonal VIs, though the different VIs varied in their predictive strength across sites and plant functional types. Widely used NDVI and EVI exhibited limited potential in tracking growing season phenology of ENF ecosystems, while indices sensitive to water (i.e., LSWI) or less influenced by soil (i.e., OSAVI) may have unrevealed powers in indicating phenological transitions. OSAVI was shown to be a strong predictor of the end of the growing season in ENF ecosystems, suggesting that this VI may offer a new strategy for modeling the phenology of ENF sites. We conclude that combinations of multiple indices may improve the remote sensing of land surface phenology, as evidenced by the good agreement between modeled and observed growing season transitions and its length in our evaluation.