Abstract: The satellite measurements are the contribution of surface reflectance and atmospheric back scattering, and it must distinguish one from the other for the quantitative remote sensing require the actual surface reflectance information. For the sake of solving this problem, a fast atmospheric correction algorithm based on Landsat-TM imagery is proposed in this paper. An outstanding feacture of this method is simple, practicable and easy to be used. Based on the traditional dark-object subtraction (DOS) technique, according to the reasonable analyse on dark-object and associated with the atmospheric radiative transfer models, the paper improves the determinate measure of atmospheric path radiance. In the paper, the author regards the vegetation in the shadow in a mountainous area as dark-object because the vegetation reflectance is very weak in the visible light and the direct daylight illumination in shadow area is also weak. Therefore, we can suppose that the vegetation reflectance in shadow area is zero approximatively so that the pixel value that is not zero all the same ought to arise from the atmospheric back scattering. Depending on the hypothesis we can calculate atmospheric path radiance and deduce the molecule optical thickness and aerosol optical thickness from the atmospheric correction algorithm. And then the parameters of atmospheric correction could be identified so that it can retrieve the surface reflectance. But this result does not answer to our precision request since the vegetation reflectance in the shadow is not zero in fact, especially in the near infrared. And for removing the error and increasing the precision the author emendates the atmospheric path radiance by the overlap way. The material approach is that we choose the exuberance and direct-daylight illumination vegetation area around the aforementioned vegetation area in the mountain shadow and regard its value as the new actual vegetation reflectance and process the second operation. By means of the overlap way we can discover the second result is more close to the actual surface reflectance. Of course, it can operate time after time until the result answer to the precision request by this method. At the same time the operate isn’t complexity by dint of the programme.At the end of this paper, as one of its exemplifications, we applied the fast atmospheric correction algorithm to Landsat-5 TM imagery obtained on December 8, 1998 to retrieve the surface reflectance, and compared the computational result with the original reflectance and field-measured reflectance. As a result, it shows that the computational reflectance by the method almost tallies with the field-measured reflectance. In addition, due to eliminating the atmospheric scatter influence the edges of the surface objects are distinct and easy to identify. So the imagery quality have been reformed too. In conclusion, based on this fast atmospheric correction algorithm, it can retrieve the actual surface reflectance from the imagery itself because all needful parameters, such as atmospheric optical thickness, single scattering albedo, are derived from the information contained in the imagery. This method adapt well to the area where the landform is plainness relatively and when it is sunshiny.