The spectral characteristic of remotely sensed image is mainly the results of integrative effects on spectrum from heterogeneous ground reflectors. Investigating its spatial distribution characteristics may be helpful for image interpreting and modeling based on remote sensing technique. In the present study, spatial heterogeneity of remotely sensed multispectral TM image across a hilly area in late October was studied by the combination of statistical method and multifractal analysis. The results showed that distribution of digital number (DN) values of visible spectra (0.45～0.69 μm) had statistical scale-invariance. The generalized fractal dimension function Dq suggested that distribution of TM 2 (0.52～0.60 μm) DN values was monofractal type, whereas DN values of TM 1 (0.45～0.52 μm) and TM 3 (0.63～0.69 μm) had multifractal distribution characteristics. The parameters (αmax-αmin) and ［f(αmax)-f(αmin)］ of multifractal spectra further indicated that TM 3 DN values had the highest spatial heterogeneity and most abundant information, followed by TM 1, while the extremely narrow spectrum of TM 2 DN values showed its relatively low spatial heterogeneity and information capacity.