The moderate resolution imaging spectroradiometer (MODIS) sensor contains 36 bands at spatial resolutions of 250 m (e.g., bands 1-2), 500 m (e.g., bands 3-7), and 1000 m (e.g., bands 8-36). The first seven bands covering the visible to midinfrared wavelengths have been used widely for monitoring the Earth's surface. However, 500 m MODIS bands 3-7 present challenges for use in land cover/land use applications, as many land cover features on the Earth's surface possess complex structures with a spatial resolution finer than 500 m. Fusing MODIS 250 m bands 1-2 and 500 m bands 3-7 is an attractive proposition, that is, increasing the spatial resolution of bands 3-7. The geostatistical based downscaling approach, area-to-point regression kriging (ATPRK), has shown great potential for MODIS image downscaling. However, it considers the global relationship between bands 1-2 and each of bands 3-7 to select a 250 m PAN-like band from bands 1-2, which may not take full advantage of both bands 1 and 2. In this paper, a new geostatistical downscaling method of spectral-spatial adaptive ATPRK (SSAATPRK) is proposed for MODIS image downscaling. Both fine spatial resolution bands (i.e., bands 1 and 2) are used as the input to SSAATPRK, and there is no need to choose a PAN-like band for each coarse band, as in the original ATPRK method. SSAATPRK was compared to four benchmark image fusion methods, including principal component analysis, high-pass filtering, ATPRK, and adaptive ATPRK (AATPRK), using one synthetic MODIS image experiment and two real MODIS image experiments. Both visual and quantitative evaluations demonstrated that SSAATPRK produced results consistently with the greatest amount of spatial detail and the largest accuracy. Furthermore, SSAATPRK inherits completely the advantages of ATPRK and AATPRK, while extending them for MODIS image downscaling.