Vegetation phenology is the chronology of periodic phases of development. It constitutes an efficient bio-indicator of impacts of climatechanges and a key parameter for understanding and modelling vegetation-climate interactions and their implications on carbon cycling. Numerousstudies were devoted to the remote sensing of vegetation phenology. Most of these were carried out using data acquired by AVHRR instrumentonboard NOAA meteorological satellites. Since 1999, multispectral images were acquired over the whole earth surface every one to two days byMODIS instrument onboard Terra and Aqua platforms. In comparison with AVHRR, MODIS constitutes a significant technical improvement interms of spatial resolution, spectral resolution, geolocation accuracy, atmospheric corrections scheme and cloud screening and sensor calibration.In this study, 250 m daily MODIS data were used to derive precise vegetation phenological dates over deciduous forest stands. Phenologicalmarkers derived from MODIS time-series and provided by MODIS Global Land Cover Dynamics product (MOD12Q2) were compared to fieldmeasurements carried out over the main deciduous forest stands across France and over five years. We show that the inflexion point of theasymmetric double-sigmoid function fitted to NDVI temporal profile is a good marker of the onset of green-up in deciduous stands. At plot level,the prediction uncertainty is 8.5 days and the bias is 3.5 days. MODIS Global Land Cover Dynamics MOD12Q2 provides estimates of onset ofgreen-up dates which deviate substantially from in situ observations and do not perform better than the null model. RMSE values are 20.5 days(bias -17 days) using the onset of greenness increase and 36.5 days (bias 34.5 days) using the onset of greenness maximum. An improvement ofprediction quality is obtained if we consider the average of MOD12Q2 onset of greenness increase and maximum as marker of onset of green-update. RMSE decreases to 16.5 days and bias to 7.5 days.