The present invention provides a method for detecting a near-surface average temperature based on MODIS data. The method comprises: A. obtaining an EVI' based on the MODIS data, the EVI' being an EVI of eight days; B. obtaining an EVIc according to a near-surface average temperature actually measured by a weather station and a corresponding LST and a corresponding EVI in the MODIS data, the EVIc being an EVI of full vegetation coverage; and C. calculating a near-surface temperature Ta according to a formula 8 which is defined as in the description, wherein the Ts<day> and the Ts<night> are obtained based on the MODIS data. Compared with statistical methods, split-window algorithms and thermodynamic methods, the algorithm of the invention requires only one undetermined parameter EVIc, and the module is simple and easy to use, providing good expansibility. Besides, compared with the temperature-vegetation index method, the algorithm of the invention needs not to open up a calculation window, so that the self-similarity of near-surface temperature inversion results is reduced, and the calculation efficiency is improved; what's more, the influence caused by poor effects of Ts-NDVI linear fitting in the calculation window is prevented.