In this paper, accuracy analysis and error correction for MODIS reflectance product are accomplished based on ground measurement data. Some ground measurement points (mixed pixels) are selected in MODIS multi-spectral remote sensing image at the eastern Beach of Chongming Island, Shanghai, China. In order to measure surface reflectance of these mixed pixels, registration is first made between MODIS and QuickBird images, then the same sizable areas to one MODIS pixel are chose in QuickBird image (a pixel for MODIS is equal to many pixels for QuickBird), and image classification is made in each QuikBird sample area. Finally, taking the ratio of every class area to the whole area as weight, the reflectance is measured of every ground object, and the weighted mean reflectance is calculated of the whole area (a pixel for MODIS). Regarded the weighted mean reflectance as the measurement reflectance of corresponding pixel in MODIS image, the measurement reflectance of each pixel is compared with the MODIS retrieval reflectance, the accuracy of MODIS reflectance products is analyzed, the relation between MODIS retrieval reflectance and measurement reflectance is gained through linear regression by which an error correction are made to MODIS reflectance product. The comparison between errors corrected MODIS reflectance and the reflectance of ground checking point shows that the method in this paper can improve the accuracy of MODIS surface reflectance product efficaciously.