By assuming that the ground reflectance is constant, variations of satellite signal may be attributed to variations of the atmospheric optical properties. Based upon this, Tanre et al, 1988 has defined the structure function for deriving aerosol optical depth from TM images. Holben (1992) also has gotten the similar results using AVHRR data. Basically, the structure function is defined by single direction only. It can't represent the complete properties of the surface structure. To overcome this, a multi directional structure function, which should include more information than the single directional from the surface structure, is used in this paper to improve the structure function method for SPOT data. In general, the error of aerosol optical depth determination could be enlarged by the different pattern of structure function distribution which is caused by the characteristics of local landcovers, terrain and the high resolution property of SPOT data. In order to reduce such error, it is necessary to determine an optimum number of distance to truncate the different pattern of structure function distribution. After these improvements, the results of this study are reasonable when compared with the Dense Dark Vegetation method's results.